PageSourceSearch

https://jhalderm.com/papers/

html jhalderm.com collected 2026-09-24 08:38:11 UTC 177,779 bytes, 2,485 lines download raw bytes

1<!DOCTYPE html>
2<html lang="en">
3<head>
4  <meta charset="utf-8">
5  <meta name="description" content="J. Alex Halderman - Publications">
6  <meta name="viewport" content="width=device-width, initial-scale=1.0">
7  <title>J. Alex Halderman - Publications</title>
8  <link rel="stylesheet" href="../home/common.css">
9  <style>
10    .toggler { font-family: helvetica, arial, sans-serif; font-size: 12px; color: #aaa; }
11    .toggler:hover { color: #4A78BA; }
12    .abstract, .spacer { margin: 0 0 18px 0; padding: 12px 1px 1px 1px; }
13    .abstract > a.toggler { position: absolute; padding-left: 13px; margin-top: -7.28px; }
14    .toggleOff > a.toggler { background: url(../home/plus.png) center left no-repeat; }
15    .toggleOn > a.toggler { background: url(../home/minus.png) center left no-repeat; }
16    .abstract > p { display: none; font-size: 91%; margin: 8px 0 0 13px; text-align: justify; }
17    .toggleOn > p { display: block; }
18    .title { margin-bottom: 2px; }
19    .title > a { font-weight: bold; }
20    .altlink { margin-left: 6px; }
21    .tt { font-family: monospace; }
22    #expandall { height: 20px; margin: 3px 0; }
23    @media screen { h3#top { margin-top: 0; } }
24    @media print { .toggler, .altlink, #expandall { display: none; } }
25  </style>
26  
26<script>
27    document.addEventListener('DOMContentLoaded', () => {
28      const abstracts = Array.from(document.getElementsByClassName('abstract'));
29
30      function makeToggler(text, onClick) {
31        const a = document.createElement('a');
32        a.className = 'toggler';
33        a.href = '#';
34        a.textContent = text;
35        a.addEventListener('click', (e) => { e.preventDefault(); onClick(); });
36        return a;
37      }
38
39      function setOpen(div, open) {
40        div.classList.toggle('toggleOn', open);
41        div.classList.toggle('toggleOff', !open);
42      }
43
44      function allOpen() {
45        return abstracts.every((div) => div.classList.contains('toggleOn'));
46      }
47
48      const expandAll = makeToggler('Expand all abstracts', () => {
49        const open = !allOpen();
50        abstracts.forEach((div) => setOpen(div, open));
51        updateExpandAll();
52      });
53      document.getElementById('expandall').appendChild(expandAll);
54
55      function updateExpandAll() {
56        expandAll.textContent = allOpen() ? 'Close all abstracts' : 'Expand all abstracts';
57      }
58
59      abstracts.forEach((div) => {
60        setOpen(div, false);
61        div.prepend(makeToggler('Abstract', () => {
62          setOpen(div, !div.classList.contains('toggleOn'));
63          updateExpandAll();
64        }));
65      });
66    });
67  </script>
67
68</head>
69<body>
70
71  <div class="fore">
72    <h1>Publications</h1>
73    <h2>
74      <span class="morelinks">
75        <a href="https://scholar.google.com/citations?user=h6yXnyEAAAAJ">&raquo;&nbsp;<span class="wide">Google Scholar</span><span class="narrow">Scholar</span></a>
76        <a href="../">&raquo;&nbsp;home</a>
77      </span>
78      <a href="../">J. Alex Halderman</a>
79    </h2>
80
81    <div id="expandall"></div>
82
83    <h3 id="top">Peer-reviewed research papers</h3>
84
85    <div class="citation">
86      <div class="title"><a href="https://www.usenix.org/system/files/usenixsecurity26-ablove.pdf">Technical Analysis of the Geedge Networks Firewall Source Code Leak</a></div>
87      <div class="author">Anna&nbsp;Ablove, Johnnie&nbsp;Walker, Ben&nbsp;Wolin, Niklas&nbsp;Niere, Felix&nbsp;Lange, Aaron&nbsp;Ortwein, Armin&nbsp;Huremagic, Richa&nbsp;Priyanka, Ali&nbsp;Zohaib, Jade&nbsp;Sheffey, Nico&nbsp;Heitmann, J.&nbsp;Alex&nbsp;Halderman, Juraj&nbsp;Somorovsky, Amir&nbsp;Houmansadr, Roya&nbsp;Ensafi, Mingshi&nbsp;Wu, and Eric&nbsp;Wustrow</div>
88      <div class="ref"><i>35th USENIX Security Symposium</i>, August 2026</div>
89      <div class="abstract">
90        <p>In September 2025, over 100K internal documents (including code,
91        communications, etc.) from Geedge Networks, a Chinese DPI company with
92        ties to the Great Firewall of China, were leaked to the public. In this
93        paper, we analyze the source code from this leak, focusing on Geedge
94        Networks&rsquo; flagship product, the Tiangou Secure Gateway (TSG)
95        firewall. Working across multiple repositories, we successfully build
96        and run a local copy of TSG&mdash;revealing key aspects of its
97        architecture, including the protocols it is capable of parsing and the
98        format of blocking rules used to censor sites, proxies, and other
99        resources. Finally, we extract several fingerprints from TSG, including
100        custom random number generators and parsing idiosyncrasies that allow us
101        to identify its use and similar deployments in the Great Firewall of
102        China.</p>
103        <p>This is the first time that the source code of a commercial DPI has
104        been leaked, and our work is the first code analysis of a core firewall
105        component used in national censorship infrastructure. This unprecedented
106        investigation offers insights that can assist circumvention developers
107        and Internet security researchers in further understanding the
108        capabilities and limitations of modern censorship technology.</p>
109      </div>
110    </div>
111
112    <div class="citation">
113      <div class="title"><a href="https://pubsonline.informs.org/doi/full/10.1287/opre.2023.0422">Improving the Security of United States Elections with Robust Optimization</a> &mdash; <b>Harvey J. Greenberg Research Award</b></div>
114      <div class="author">Braden&nbsp;Crimmins, J.&nbsp;Alex&nbsp;Halderman, and Bradley&nbsp;Sturt</div>
115      <div class="ref"><i>Operations Research</i> 73(1):61-85, January&ndash;February 2025</div>
116      <div class="abstract">
117        <p>For more than a century, election officials across the United States
118        have inspected voting machines before elections using a procedure called
119        logic and accuracy testing (LAT). This procedure consists of election
120        officials casting a test deck of ballots into each voting machine and
121        confirming the machine produces the expected vote total for each
122        candidate. We bring a scientific perspective to LAT by introducing the
123        first formal approach to designing test decks with rigorous security
124        guarantees. Specifically, our approach employs robust optimization to
125        find test decks that are guaranteed to detect any voting machine
126        misconfiguration that would cause votes to be swapped across candidates.
127        Of all the test decks with this security guarantee, our robust
128        optimization problem yields the test deck with the minimum number of
129        ballots, thereby minimizing implementation costs for election officials.
130        To facilitate deployment at scale, we develop a practically efficie
130nt
131        exact algorithm for solving our robust optimization problems based on
132        the cutting plane method. In partnership with the Michigan Bureau of
133        Elections, we retrospectively applied our approach to all 6,928 ballot
134        styles from Michigan&rsquo;s November 2022 general election; this
135        retrospective study reveals that the test decks with rigorous security
136        guarantees obtained by our approach require, on average, only 1.2% more
137        ballots than current practice. Our approach has since been piloted in
138        real-world elections by the Michigan Bureau of Elections as a low-cost
139        way to improve election security and increase public trust in democratic
140        institutions.</p>
141      </div>
142    </div>
143
144    <div class="citation">
145      <div class="title"><a href="https://dl.acm.org/doi/pdf/10.1145/3646547.3689012">Ten Years of ZMap</a></div>
146      <div class="author">Zakir&nbsp;Durumeric, David&nbsp;Adrian, Phillip&nbsp;Stephens, Eric&nbsp;Wustrow, and J.&nbsp;Alex&nbsp;Halderman</div>
147      <div class="ref"><i>24th ACM Internet Measurement Conference</i> (IMC), November 2024</div>
148      <div class="abstract">
149        <p>Since ZMap&rsquo;s debut in 2013, networking and security researchers
150        have used the open-source scanner to write hundreds of research papers
151        that study Internet behavior. In addition, ZMap has been adopted by the
152        security industry to build new classes of enterprise security and
153        compliance products. Over the past decade, much of ZMap&rsquo;s
154        behavior&mdash;ranging from its pseudorandom IP generation to its packet
155        construction&mdash;has evolved as we have learned more about how to scan
156        the Internet. In this work, we quantify ZMap&rsquo;s adoption over the
157        ten years since its release, describe its modern behavior (and the
158        measurements that motivated changes), and offer lessons from releasing
159        and maintaining ZMap for future tools.</p>
160      </div>
161    </div>
162
163    <div class="citation">
164      <div class="title"><a href="https://www.usenix.org/system/files/usenixsecurity24-crimmins.pdf">DVSorder: Ballot Randomization Flaws Threaten Voter Privacy</a> <span class="altlink">(<a href="https://dvsorder.org">website and tools</a>)</span></div>
165      <div class="author">Braden&nbsp;L.&nbsp;Crimmins, Dhanya&nbsp;Y.&nbsp;Narayanan, Drew&nbsp;Springall, and J.&nbsp;Alex&nbsp;Halderman</div>
166      <div class="ref"><i>33rd USENIX Security Symposium</i>, August 2024 &mdash; <b>Best&nbsp;Paper&nbsp;Award</b></div>
167      <div class="abstract">
168        <p>A trend towards publishing ballot-by-ballot election results has
169        created new risks to voter privacy due to inadequate protections by
170        election technology. These risks are manifested by a vulnerability we
171        discovered in precinct-based ballot scanners made by Dominion Voting
172        Systems, which are used in parts of 21 states and Canada. In a variety
173        of scenarios, the flaw&mdash;which we call DVSorder&mdash;would allow
174        attackers to link individuals with their votes and compromise ballot
175        secrecy. The root cause is that the scanners assign pseudorandom ballot
176        identifiers using a linear congruential generator, an approach known
177        since the 1970s to be insecure. Dominion attempted to obfuscate the
178        generator&rsquo;s output, but we show that it can be broken using only
179        pen and paper to reveal the order in which all ballots were cast. Unlike
180        past ballot randomization flaws, which typically required insider access
181        to exploit or access to proprietary software to discover, DVSorder can
182        be discovered and exploited using only public information.</p>
183        <p>In addition, the election sector&rsquo;s response to our findings
184        provides a case study highlighting gaps in regulations and vulnerability
185        management within this area of critical infrastructure. Although
186        Dominion released a software update in response to DVSorder, some
187        localities have continued to publish vulnerable data due to inadequate
188        information sharing and mitigation planning, and at least one state has
189        deferred addressing the flaw until after the 2024 presidential election,
190        more than two years following our disclosure.</p>
191      </div>
192    </div>
193
194    <div class="citation">
195      <div class="title"><a href="https://www.petsymposium.org/foci/2024/foci-2024-0003.pdf">Just add WATER: WebAssembly-based Circumvention Transports</a></div>
196      <div class="author">Erik&nbsp;Chi, Gaukas&nbsp;Wang, J.&nbsp;Alex&nbsp;Halderman, Eric&nbsp;Wustrow, and Jack&nbsp;Wampler</div>
197      <div class="ref"><i>Workshop on Free and Open Communications on the Internet (FOCI)</i>, February 2024</div>
198      <div class="abstract">
199        <p>As Internet censors rapidly evolve new blocking techniques,
200        circumvention tools must also adapt and roll out new strategies to
201        remain unblocked. But new strategies can be time consuming for
202        circumventors to develop and deploy, and usually an update to one tool
203        often requires significant additional effort to be ported to others.
204        Moreover, distributing the updated application across different
205        platforms poses its own set of challenges.</p>
206        <p>In this paper, we introduce WATER (WebAssembly Transport Executables
207        Runtime), a novel design that enables applications to use a
208        WebAssembly-based application-layer to wrap network transports (e.g.,
209        TLS). Deploying a new circumvention technique with WATER only requires
210        distributing the WebAssembly Transport Module (WATM) binary and any
211        transport-specific configuration, allowing dynamic transport updates
212        without any change to the application itself. WATMs are also designed to
213        be generic such that different applications using WATER can use the same
214        WATM to rapidly deploy successful circumvention techniques to their own
215        users, facilitating rapid interoperability between independent
216        circumvention tools.</p>
217      </div>
218    </div>
219
220    <div class="citation">
221      <div class="title"><a href="https://arxiv.org/pdf/2107.10344">Challenges in Cybersecurity: Lessons from Biological Defense Systems</a></div>
222      <div class="author">Edward&nbsp;Schrom, Ann&nbsp;Kinzig, Stephanie&nbsp;Forrest, Andrea&nbsp;L.&nbsp;Graham, Simon&nbsp;A.&nbsp;Levin, Carl&nbsp;T.&nbsp;Bergstrom, Carlos&nbsp;Castillo-Chavez, James&nbsp;P.&nbsp;Collins, Rob&nbsp;J.&nbsp;de&nbsp;Boer, Adam&nbsp;Doup&eacute;, Roya&nbsp;Ensafi, Stuart&nbsp;Feldman, Bryan&nbsp;T.&nbsp;Grenfell, J.&nbsp;Alex&nbsp;Halderman, Silvie&nbsp;Huijben, Carlo&nbsp;Maley, Melanie&nbsp;Moses, Alan&nbsp;S.&nbsp;Perelson, Charles&nbsp;Perrings, Joshua&nbsp;Plotkin, Jennifer&nbsp;Rexford, and Mohit&nbsp;Tiwari</div>
223      <div class="ref"><i>Mathematical Biosciences</i>, vol. 362, August 2023</div>
224      <div class="abstract">
225        <p>We explore the commonalities between methods for assuring the
226        security of computer systems (cybersecurity) and the mechanisms that
227        have evolved through natural selection to protect vertebrates against
228        pathogens, and how insights derived from studying the evolution of
229        natural defenses can inform the design of more effective cybersecurity
230        systems. More generally, security challenges are crucial for the
231        maintenance of a wide range of complex adaptive systems, including
232        financial systems, and again lessons learned from the study of the
233        evolution of natural defenses can provide guidance for the protection of
234        such systems.</p>
235      </div>
236    </div>
237
238    <div class="citation">
239      <div class="title"><a href="https://arxiv.org/pdf/2207.14394.pdf">Logic and Accuracy Testing: A Fifty-State Review</a></div>
240      <div class="author">Josiah&nbsp;Walker, Nakul&nbsp;Bajaj, Braden&nbsp;L.&nbsp;Crimmins, and J.&nbsp;Alex&nbsp;Halderman</div>
241      <div class="ref"><i>7th International Joint Conference on Electronic Voting</i> (E-Vote-ID &rsquo;22), October 2022</div>
242      <div class="abstract">
243        <p>Pre-election logic and accuracy (L&amp;A) testing is a process in
244        which election officials validate the behavior of voting equipment by
245        casting a known set of test ballots and confirming the expected results.
246        Ideally, such testing can serve to detect certain forms of human error
247        or fraud and help bolster voter confidence. We present the first
248        detailed analysis of L&amp;A testing practices across the United States.
249        We find that while all states require L&amp;A testing before every
250        election, their implementations vary dramatically in scope,
251        transparency, and rigorousness. We summarize each state&rsquo;s
252        requirements and score them according to uniform criteria. We also
253        highlight best practices and flag opportunities for improvement, in
254        hopes of encouraging broader adoption of more effective L&amp;A
255        processes.</p>
256      </div>
257    </div>
258
259    <div class="citation">
260      <div class="title"><a href="https://www.usenix.org/system/files/sec22-halderman.pdf">The Antrim County 2020 Election Incident: An Independent Forensic Investigation</a> <span class="altlink">(<a href="https://www.michigan.gov/documents/sos/Antrim_720623_7.pdf">original expert report</a>)</span></div>
261      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
262      <div class="ref"><i>31st USENIX Security Symposium</i>, August 2022 &mdash; <b>Best&nbsp;Paper&nbsp;Award</b></div>
263      <div class="abstract">
264        <p>In November 2020, Antrim County, Michigan published unofficial
265        election results that misstated totals in the presidential race and
266        other contests by up to several thousand votes. Antrim subsequently
267        issued a series of corrections, and the certified presidential results
268        were confirmed by a hand count. Nevertheless, Antrim was cited by the
269        President of the United States as evidence of widespread fraud, and it
270        remains a centerpiece of conspiracy theories about the 2020 election. At
271        the request of the Michigan Secretary of State and Attorney General, I
272        performed a forensic investigation of the incident. Using data from the
273        election system, I precisely reproduce the major anomalies, explain
274        their cause, and verify that they have been corrected. I also uncover
275        other errors affecting specific down-ballot contests that have
276        <i>not</i> been corrected, despite the unusual attention focused on the
277        results, one of which may have changed the outcome of a local contest.
278        Based on this analysis, I refute misinformation about the incident,
279        concluding that it was not the result of a security breach but rather a
280        series of operator errors compounded by inadequate procedures and
281        insufficiently defensive software design. These events offer lessons for
282        improving election administration and highlight the value of rigorously
283        investigating election technology incidents for enhancing accuracy and
284        public trust.</p>
285      </div>
286    </div>
287
288    <div class="citation">
289      <div class="title"><a href="https://www.usenix.org/system/files/sec22fall_xue-diwen.pdf">OpenVPN is Open to VPN Fingerprinting</a></div>
290      <div class="author">Diwen&nbsp;Xue, Reethika&nbsp;Ramesh, Arham&nbsp;Jain, Michalis&nbsp;Kallitsis, J.&nbsp;Alex&nbsp;Halderman, Jedidiah&nbsp;R.&nbsp;Crandall, and Roya&nbsp;Ensafi</div>
291      <div class="ref"><i>31st USENIX Security Symposium</i>, August 2022 &mdash; <b>Best&nbsp;Paper&nbsp;Award</b> &mdash; <b>Internet&nbsp;Defense&nbsp;Prize Winner</b></div>
292      <div class="abstract">
293        <p>VPN adoption has seen steady growth over the past decade due to
294        increased public awareness of privacy and surveillance threats. In
295        response, certain governments are attempting to restrict VPN access by
296        identifying connections using &ldquo;dual use&rdquo; DPI technology. To
297        investigate the potential for VPN blocking, we develop mechanisms for
298        accurately fingerprinting connections using OpenVPN, the most popular
299        protocol for commercial VPN services. We identify three fingerprints
300        based on protocol features such as byte pattern, packet size, and server
301        response. Playing the role of an attacker who controls the network, we
302        design a two-phase framework that performs passive fingerprinting and
303        active probing in sequence. We evaluate our framework in partnership
304        with a million-user ISP and find that we identify over 85% of OpenVPN
305        flows with only negligible false positives, suggesting that
306        OpenVPN-based services can be effectively blocked with little collateral
307        damage. Although some commercial VPNs implement countermeasures to avoid
308        detection, our framework successfully identified connections to 34 out
309        of 41 &ldquo;obfuscated&rdquo; VPN configurations. We discuss the
310        implications of the VPN fingerprintability for different threat models
311        and propose short-term defenses. In the longer term, we urge commercial
312        VPN providers to be more transparent about their obfuscation approaches
313        and to adopt more principled detection countermeasures, such as those
314        developed in censorship circumvention research.</p>
315      </div>
316    </div>
317
318    <div class="citation">
319      <div class="title"><a href="https://arxiv.org/pdf/2111.08662.pdf">RemoteVote and SAFE Vote: Towards Usable End-to-End Verification for Vote-by-Mail</a></div>
320      <div class="author">Braden&nbsp;L.&nbsp;Crimmins, Marshall&nbsp;Rhea, and J.&nbsp;Alex&nbsp;Halderman</div>
321      <div class="ref"><i>7th Workshop on Advances in Secure Electronic Voting</i>, February 2022</div>
322      <div class="abstract">
323        <p>Postal voting is growing rapidly in the U.S., with 43% of voters
324        casting ballots by mail in 2020, yet until recently there has been
325        little research about extending the protections of end-to-end verifiable
326        (E2E-V) election schemes to vote-by-mail contexts. The first&mdash;and
327        to date, only&mdash;framework to focus on this setting is STROBE, which
328        has important usability limitations. In this work, we present two
329        approaches, RemoteVote and SAFE Vote, that allow mail-in voters to
330        benefit from E2E-V without changing the voter experience for those who
331        choose not to participate in verification. To evaluate these systems and
332        compare them with STROBE, we consider an expansive set of properties,
333        including novel attributes of usability and verifiability, several of
334        which have applicability beyond vote-by-mail contexts. We hope that our
335        work will help catalyze further progress towards universal applicability
336        of E2E-V for real-world elections.</p>
337      </div>
338    </div>
339
340    <div class="citation">
341      <div class="title"><a href="/pub/papers/ballotscanning-evoteid21.pdf">Improving the Accuracy of Ballot Scanners Using Supervised Learning</a></div>
342      <div class="author">Sameer&nbsp;Barretto, William&nbsp;Chown, David&nbsp;Meyer, Aditya&nbsp;Soni, Atreya&nbsp;Tata, and J.&nbsp;Alex&nbsp;Halderman</div>
343      <div class="ref"><i>6th International Joint Conference on Electronic Voting</i> (E-Vote-ID &rsquo;21), October 2021</div>
344      <div class="abstract">
345        <p>Most U.S. voters cast hand-marked paper ballots that are counted by
346        optical scanners. Deployed ballot scanners typically utilize simplistic
347        mark-detection methods, based on comparing the measured intensity of
348        target areas to preset thresholds, but this technique is known to
349        sometimes misread &ldquo;marginal&rdquo; marks that deviate from ballot
350        instructions. We investigate the feasibility of improving scanner
351        accuracy using supervised learning. We train a convolutional neural
352        network to classify various styles of marks extracted from a large
353        corpus of voted ballots. This approach achieves higher accuracy than a
354        naive intensity threshold while requiring far fewer ballots to undergo
355        manual adjudication. It is robust to imperfect feature extraction, as
356        may be experienced in ballots that lack timing marks, and efficient
357        enough to be performed in real time using contemporary central-count
358        scanner hardware.</p>
359      </div>
360    </div>
361
362    <div class="citation">
363      <div class="title"><a href="/pub/papers/omniballot-sec21.pdf">Security Analysis of the Democracy Live Online Voting System</a> <span class="altlink">(<a href="/pub/papers/omniballot-20.pdf">initial public version</a>)</span></div>
364      <div class="author">Michael&nbsp;A.&nbsp;Specter and J.&nbsp;Alex&nbsp;Halderman</div>
365      <div class="ref"><i>30th USENIX Security Symposium</i>, August 2021</div>
366      <div class="abstract">
367        <p>Democracy Live&rsquo;s OmniBallot platform is a web-based system for
368        blank ballot delivery, ballot marking, and online voting. In early 2020,
369        three states&mdash;Delaware, West Virginia, and New
370        Jersey&mdash;announced that they would allow certain voters to cast
371        votes online using OmniBallot, but, despite the well established risks
372        of Internet voting, the system has never before undergone a public,
373        independent security review.</p>
374        <p>We reverse engineered the client-side portion of OmniBallot, as used
375        in Delaware, in order to detail the system&rsquo;s operation and analyze
376        its security. We find that OmniBallot uses a simplistic approach to
377        Internet voting that is vulnerable to vote manipulation by malware on
378        the voter&rsquo;s device and by insiders or other attackers who can
379        compromise Democracy Live, Amazon, Google, or Cloudflare. In addition,
380        Democracy Live, which had no privacy policy prior to our work, receives
381        sensitive personally identifiable information&mdash;including the
382        voter&rsquo;s identity, ballot selections, and browser
383        fingerprint&mdash;that could be used to target political ads or
384        disinformation campaigns. Even when OmniBallot is used to mark ballots
385        that will be printed and returned in the mail, the softw
385are sends the
386        voter&rsquo;s identity and ballot choices to Democracy Live, an
387        unnecessary risk that jeopardizes the secret ballot.</p>
388        <p>We recommend changes to make the platform safer for ballot delivery
389        and marking. However, we conclude that using OmniBallot for electronic
390        ballot return represents a severe risk to election security and could
391        allow attackers to alter election results without detection. In response
392        to our findings, Delaware and New Jersey halted their use of OmniBallot
393        for online voting, but it remains available in other jurisdictions, as
394        do similar tools that likely face the same serious risks.</p>
395      </div>
396    </div>
397
398    <div class="citation">
399      <div class="title"><a href="https://dl.acm.org/doi/pdf/10.1145/3419394.3423665">Investigating Large-Scale HTTPS Interception in Kazakhstan</a></div>
400      <div class="author">Ram&nbsp;Sundara&nbsp;Raman, Leonid&nbsp;Evdokimov, Eric&nbsp;Wustrow, J.&nbsp;Alex&nbsp;Halderman, and Roya&nbsp;Ensafi</div>
401      <div class="ref"><i>20th ACM Internet Measurement Conference</i> (IMC), October 2020</div>
402      <div class="abstract">
403        <p>Increased adoption of HTTPS has created a largely encrypted web, but
404        these security gains are on a collision course with governments that
405        desire visibility into and control over user communications. Last year,
406        the government of Kazakhstan conducted an unprecedented large-scale
407        HTTPS interception attack by forcing users to trust a custom root
408        certificate. We were able to detect the interception and monitor its
409        scale and evolution using measurements from in-country vantage points
410        and remote measurement techniques. We find that the attack targeted
411        connections to 37 unique domains, with a focus on social media and
412        communication services, suggesting a surveillance motive, and that it
413        affected a large fraction of connections passing through the
414        country&rsquo;s largest ISP, Kazakhtelecom. Our continuous real-time
415        measurements indicated that the interception system was shut down after
416        being intermittently active for 21 days. Subsequently, supported by our
417        findings, two major browsers (Mozilla Firefox and Google Chrome)
418        completely blocked the use of Kazakhstan&rsquo;s custom root. However,
419        the incident sets a dangerous precedent, not only for Kazakhstan but for
420        other countries that may seek to circumvent encryption online.</p>
421      </div>
422    </div>
423
424    <div class="citation">
425      <div class="title"><a href="/pub/papers/refraction-pets20.pdf">Running Refraction Networking for Real</a></div>
426      <div class="author">Benjamin&nbsp;VanderSloot, Sergey&nbsp;Frolov, Jack&nbsp;Wampler, Sze&nbsp;Chuen&nbsp;Tan, Irv&nbsp;Simpson, Michalis&nbsp;Kallitsis, J.&nbsp;Alex&nbsp;Halderman, Nikita&nbsp;Borisov, and Eric&nbsp;Wustrow</div>
427      <div class="ref"><i> 20th Privacy Enhancing Technologies Symposium</i> (PETS), July 2020</div>
428      <div class="abstract">
429        <p>Refraction networking is a next-generation censorship circumvention
430        approach that locates proxy functionality in the network itself, at
431        participating ISPs or other network operators. Following years of
432        research and development and a brief pilot, we established the
433        world&rsquo;s first production deployment of a Refraction Networking
434        system. Our deployment uses a high-performance implementation of the
435        TapDance protocol and is enabled as a transport in the popular
436        circumvention app Psiphon. It uses TapDance stations at four physical
437        uplink locations of a mid-sized ISP, Merit Network, with an aggregate
438        bandwidth of 140 Gbps. By the end of 2019, our system was enabled as a
439        transport option in 559,000 installations of Psiphon, and it served
440        upwards of 33,000 unique users per month. This paper reports on our
441        experience building the deployment and operating it for the first year.
442        We describe how we overcame engineering challenges, present detailed
443        performance metrics, and analyze how our system has responded to dynamic
444        censor behavior. Finally, we review lessons learned from operating this
445        unique artifact and discuss prospects for further scaling Refraction
446        Networking to meet the needs of censored users.</p>
447      </div>
448    </div>
449
450    <div class="citation">
451      <div class="title"><a href="/pub/papers/slowdown-sigmetrics20.pdf">Characterizing Transnational Internet Performance and the Great Bottleneck of China</a></div>
452      <div class="author">Pengxiong&nbsp;Zhu, Keyu&nbsp;Man, Zhongjie&nbsp;Wang, Zhiyun&nbsp;Qian, Roya&nbsp;Ensafi, J.&nbsp;Alex&nbsp;Halderman, and Haixin&nbsp;Duan</div>
453      <div class="ref"><i>ACM SIGMETRICS 2020</i>, June 2020</div>
454      <div class="abstract">
455        <p>Transnational Internet performance is an important indication of a
456        country&rsquo;
456s level of infrastructure investment, globalization, and
457        openness. We conduct a large-scale measurement study of transnational
458        Internet performance in and out of 29 countries and regions, and find
459        six countries that have surprisingly low performance. Five of them are
460        African countries and the last is mainland China, a significant outlier
461        with major discrepancies between downstream and upstream performance. We
462        then conduct a comprehensive investigation of the unusual transnational
463        Internet performance of mainland China, which we refer to as the
464        &ldquo;Great Bottleneck of China&rdquo;. Our results show that this
465        bottleneck is widespread, affecting 79% of the receiver&ndash;sender
466        pairs we measured. More than 70% of the pairs suffer from extremely slow
467        speed (less than 1 Mbps) for more than 5 hours every day. In most tests
468        the bottleneck appeared to be located deep inside China, suggesting poor
469        network infrastructure to handle transnational traffic. The phenomenon
470        has far-reaching implications for Chinese users&rsquo; browsing habits
471        as well as for the ability of foreign Internet services to reach Chinese
472        customers.</p>
473      </div>
474    </div>
475
476    <div class="citation">
477      <div class="title"><a href="/pub/papers/bmd-verifiability-sp20.pdf">Can Voters Detect Malicious Manipulation of Ballot Marking Devices?</a></div>
478      <div class="author">Matthew&nbsp;Bernhard, Allison&nbsp;McDonald, Henry&nbsp;Meng, Jensen&nbsp;Hwa, Nakul&nbsp;Bajaj, Kevin&nbsp;Chang, and J.&nbsp;Alex&nbsp;Halderman</div>
479      <div class="ref"><i>41st IEEE Symposium on Security and Privacy</i> (Oakland &rsquo;20), May 2020 &mdash; <b>Best&nbsp;Student&nbsp;Paper&nbsp;Award</b></div>
480      <div class="abstract">
481        <p>Ballot marking devices (BMDs) allow voters to select candidates on a
482        computer kiosk, which prints a paper ballot that the voter can review
483        before inserting it into a scanner to be tabulated. Unlike paperless
484        voting machines, BMDs provide voters an opportunity to verify an
485        auditable physical record of their choices, and a growing number of U.S.
486        jurisdictions are adopting them for all voters. However, the security of
487        BMDs depends on how reliably voters notice and correct any adversarially
488        induced errors on their printed ballots. In order to measure
489        voters&rsquo; error detection abilities, we conducted a large study
490        (N=241) in a realistic polling place setting using real voting machines
491        that we modified to introduce an error into each printout. Without
492        intervention, only 40% of participants reviewed their printed ballots at
493        all, and only 6.6% told a poll worker something was wrong. We also find
494        that carefully designed interventions can improve verification
495        performance. Verbally instructing voters to review the printouts and
496        providing a written slate of candidates for whom to vote both
497        significantly increased review and reporting rates&mdash;although the
498        improvements may not be large enough to provide strong security in close
499        elections, especially when BMDs are used by all voters. Based on these
500        findings, we make several evidence-based recommendations to help better
501        defend BMD-based elections.</p>
502      </div>
503    </div>
504
505    <div class="citation">
506      <div class="title"><a href="/pub/papers/letsencrypt-ccs19.pdf">Let&rsquo;s Encrypt: An Automated Certificate Authority to Encrypt the Entire Web</a></div>
507      <div class="author">Josh&nbsp;Aas, Richard&nbsp;Barnes, Benton&nbsp;Case, Zakir&nbsp;Durumeric, Peter&nbsp;Eckersley, Alan&nbsp;Flores-L&oacute;pez, J.&nbsp;Alex&nbsp;Halderman, Jacob&nbsp;Hoffman-Andrews, James&nbsp;Kasten, Eric&nbsp;Rescorla, Seth&nbsp;Schoen, and Brad&nbsp;Warren</div>
508      <div class="ref"><i>26th ACM Conference on Computer and Communications Security</i> (CCS &rsquo;19), November 2019</div>
509      <div class="abstract">
510        <p>Let&rsquo;s Encrypt is a free, open, and automated HTTPS certificate
511        authority (CA) created to advance HTTPS adoption to the entire Web.
512        Since its launch in late 2015, Let&rsquo;s Encrypt has grown to become
513        the world&rsquo;s largest HTTPS CA, accounting for more currently valid
514        certificates than all other browser-trusted CAs combined. By January
515        2019, it had issued over 538 million certificates for 223 million domain
516        names. We describe how we built Let&rsquo;s Encrypt, including the
517        architecture of the CA software system (Boulder) and the structure of
518        the organization that operates it (ISRG), and we discuss lessons learned
519        from the experience. We also describe the design of ACME, the
520        IETF-standard protocol we created to automate CA-server interactions and
521        certificate issuance, and survey the diverse ecosystem of ACME clients,
522        including Certbot, a software agent we created to automate HTTPS
523        deployment. Finally, we measure Let&rsquo;s Encrypt&rsquo;s impact on
524        the Web and the CA ecosystem. We hope that the success of Let&rsquo;s
525        Encrypt can provide a model for further enhancements to the Web PKI and
526        for future Internet security infrastructure.</p>
527      </div>
528    </div>
529
530    <div class="citation">
531      <div class="title"><a href="/pub/papers/conjure-ccs19.pdf">Conjure: Summoning Proxies from Unused Address Space</a></div>
532      <div class="author">Sergey&nbsp;Frolov, Jack&nbsp;Wampler, Sze&nbsp;Chuen&nbsp;Tan, J.&nbsp;Alex&nbsp;Halderman, Nikita&nbsp;Borisov, and Eric&nbsp;Wustrow</div>
533      <div class="ref"><i>26th ACM Conference on Computer and Communications Security </i> (CCS &rsquo;19), November 2019</div>
534      <div class="abstract">
535        <p>Refraction Networking (formerly known as &ldquo;Decoy Routing&rdquo;)
536        has emerged as a promising next-generation approach for circumventing
537        Internet censorship. Rather than trying to hide individual circumvention
538        proxy servers from censors, proxy functionality is implemented in the
539        core of the network, at cooperating ISPs in friendly countries. Any
540        connection that traverses these ISPs could be a conduit for the free
541        flow of information, so censors cannot easily block access without also
542        blocking many legitimate sites. While one Refraction scheme, TapDance,
543        has recently been deployed at ISP-scale, it suffers from several
544        problems: a limited number of &ldquo;decoy&rdquo; sites in realistic
545        deployments, high technical complexity, and undesirable tradeoffs
546        between performance and observability by the censor. These challenges
547        may impede broader deployment and ultimately allow censors to block such
548        techniques.</p>
549        <p>We present Conjure, an improved Refraction Networking approach that
550        overcomes these limitations by leveraging unused address space at
551        deploying ISPs. Instead of using real websites as the decoy destinations
552        for proxy connections, our scheme connects to IP addresses where no web
553        server exists leveraging proxy functionality from the core of the
554        network. These phantom hosts are difficult for a censor to distinguish
555        from real ones, but can be used by clients as proxies. We define the
556        Conjure protocol, analyze its security, and evaluate a prototype using
557        an ISP testbed. Our results suggest that Conjure can be harder to block
558        than TapDance, is simpler to maintain and deploy, and offers
559        substantially better network performance.</p>
560      </div>
561    </div>
562
563    <div class="citation">
564      <div class="title"><a href="/pub/papers/unclear-evoteid19.pdf">UnclearBallot: Automated Ballot Image Manipulation</a></div>
565      <div class="author">Matthew&nbsp;Bernhard, Kartikeya&nbsp;Kandula, Jeremy&nbsp;Wink, and J.&nbsp;Alex&nbsp;Halderman</div>
566      <div class="ref"><i>4th International Joint Conference on Electronic Voting</i> (E-Vote-ID &rsquo;19), October 2019</div>
567      <div class="abstract">
568        <p>As paper ballots and post-election audits gain increased adoption in
569        the United States, election technology vendors are offering products
570        that allow jurisdictions to review ballot images&mdash;digital scans
571        produced by optical-scan voting machines&mdash;in their post-election
572        audit procedures. Jurisdictions including the state of Maryland rely on
573        such image audits as an alternative to inspecting the physical paper
574        ballots. We show that image audits can be reliably defeated by an
575        attacker who can run malicious code on the voting machines or election
576        management system. Using computer vision techniques, we develop an
577        algorithm that automatically and seamlessly manipulates ballot images,
578        moving voters&rsquo; marks so that they appear to be votes for the
579        attacker&rsquo;s preferred candidate. Our implementation is compatible
580        with many widely used ballot styles, and we show that it is effective
581        using a large corpus of ballot images from a real election. We also show
582        that the attack can be delivered in the form of a malicious Windows
583        scanner driver, which we test with a scanner that has been certified for
584        use in vote tabulation by the U.S. Election Assistance Commission. These
585        results demonstrate that post-election audits must inspect physical
586        ballots, not merely ballot images, if they are to strongly defend
587        against computer-based attacks on widely used voting systems.</p>
588      </div>
589    </div>
590
591    <div class="citation">
592      <div class="title"><a href="/pub/papers/https-chi19.pdf">On the Usability of HTTPS Deployment</a></div>
593      <div class="author">Matthew&nbsp;Bernhard, Jonathan&nbsp;Sharman, Claudia&nbsp;Z.&nbsp;Acemyan, Philip&nbsp;Kortum, Dan&nbsp;S.&nbsp;Wallach, and J.&nbsp;Alex&nbsp;Halderman</div>
594      <div class="ref"><i>ACM Conference on Human Factors in Computing Systems</i> (CHI &rsquo;19), May 2019</div>
595      <div class="abstract">
596        <p>HTTPS and TLS are the backbone of Internet security, however setting
597        up web servers to run these protocols is a notoriously difficult
598        process. In this paper, we perform two live subjects usability studies
599        on the deployment of HTTPS in a real-world setting. Study 1 is a within
600        subjects comparison between traditional HTTPS configuration (purchasing
601        a certificate and installing it on a server) and Let&rsquo;s Encrypt,
602        which automates much of the process. Study 2 is a between subjects study
603        looking at the same two systems, examining why users encounter usability
604        issues. Overall we confirm past results that HTTPS is difficult to
605        deploy, and we find some evidence that suggests Let&rsquo;s Encrypt is
606        an easier, more efficient method for deploying HTTPS.</p>
607      </div>
608    </div>
609
610    <div class="citation">
611      <div class="title"><a href="https://fc19.ifca.ai/voting/papers/OBHRS19.pdf">Bernoulli Ballot-Polling: A Manifest Improvement for Risk-Limiting Audits</a></div>
612      <div class="author">Kellie&nbsp;Ottoboni, Matthew&nbsp;Bernhard, J.&nbsp;Alex&nbsp;Halderman, Ronald&nbsp;L.&nbsp;Rivest, and Philip&nbsp;B&nbsp;Stark</div>
613      <div class="ref"><i>4th Workshop on Advances in Secure Electronic Voting</i> (Voting &rsquo;19), February 2019</div>
614      <div class="abstract">
615        <p>We present a method and software for ballot-polling risk-limiting
616        audits (RLAs) based on Bernoulli sampling: ballots are included in the
617        sample with probability <i>p</i>, independently. Bernoulli sampling has
618        several advantages: (1) it does not require a ballot manifest; (2) it
619        can be conducted independently at different locations, rather than
620        requiring a central authority to select the sample from the whole
621        population of cast ballots or requiring stratified sampling; (3) it can
622        start in polling places on election night, before margins are known. If
623        the reported margins for the 2016 U.S. Presidential election are
624        correct, a Bernoulli ballot-polling audit with a risk limit of 5% and a
625        sampling rate of <i>p</i><sub>0</sub>=1% would have had at least a 99%
626        probability of confirming the outcome in 42 states. (The other states
627        were more likely to have needed to examine additional ballots.)
628        Logistical and security advantages that auditing in the polling place
629        affords may outweigh the cost of examining more ballots than some other
630        methods might require.</p>
631      </div>
632    </div>
633
634    <div class="citation">
635      <div class="title"><a href="/pub/papers/geoblocking-imc18.pdf">403 Forbidden: A Global View of Geoblocking</a></div>
636      <div class="author">Allison&nbsp;McDonald, Matthew&nbsp;Bernhard, Benjamin&nbsp;VanderSloot, Will&nbsp;Scott, J.&nbsp;Alex&nbsp;Halderman, and Roya&nbsp;Ensafi</div>
637      <div class="ref"><i>18th ACM Internet Measurement Conference</i> (IMC &rsquo;18), October 2018</div>
638      <div class="abstract">
639        <p>We report the first wide-scale measurement study of serverside
640        geographic restriction, or geoblocking, a phenomenon in which server
641        operators intentionally deny access to users from particular countries
642        or regions. Many sites practice geoblocking due to legal requirements or
643        other business reasons, but excessive blocking can needlessly deny
644        valuable content and services to entire national populations.</p>
645        <p>To help researchers and policymakers understand this phenomenon, we
646        develop a semi-automated system to detect instances where whole websites
647        were rendered inaccessible due to geoblocking. By focusing on detecting
648        geoblocking capabilities offered by large CDNs and cloud providers, we
649        can reliably distinguish the practice from dynamic anti-abuse mechanisms
650        and network-based censorship. We apply our techniques to test for
651        geoblocking across the Alexa Top 10K sites from thousands of vantage
652        points in 177 countries. We then expand our measurement to a sample of
653        CDN customers in the Alexa Top 1M.</p>
654        <p>We find that geoblocking occurs across a broad set of countries and
655        sites. We observe geoblocking in nearly all countries we study, with
656        Iran, Syria, Sudan, Cuba, and Russia experiencing the highest rates.
657        These countries experience particularly high rates of geoblocking for
658        finance and banking sites, likely as a result of U.S. economic
659        sanctions. We also verify our measurements with data provided by
660        Cloudflare, and find our observations to be accurate.</p>
661      </div>
662    </div>
663
664    <div class="citation">
665      <div class="title"><a href="https://www.usenix.org/system/files/conference/usenixsecurity18/sec18-vandersloot.pdf">Quack: Scalable Remote Measurement of Application-Layer Censorship</a></div>
666      <div class="author">Benjamin&nbsp;VanderSloot, Allison&nbsp;McDonald, Will&nbsp;Scott, J.&nbsp;Alex&nbsp;Halderman, and Roya&nbsp;Ensafi</div>
667      <div class="ref"><i>27th USENIX Security Symposium</i> (Sec &rsquo;18), August 2018</div>
668      <div class="abstract">
669        <p>Remote censorship measurement tools can now detect DNS- and IP-based
670        blocking at global scale. However, a major <i>unmonitored</i> form of
671        interference is blocking triggered by deep packet inspection of
672        application-layer data. We close this gap by introducing Quack, a
673        scalable, remote measurement system that can efficiently detect
674        application-layer interference.</p>
675        <p>We show that Quack can effectively detect application-layer blocking
676        triggered on HTTP and TLS headers, and it is flexible enough to support
677        many other diverse protocols. In experiments, we test for blocking
678        across 4458 autonomous systems, an order of magnitude larger than
679        provided by country probes used by OONI. We also test a corpus of
680        100,000 keywords from vantage points in 40 countries to produce detailed
681        national blocklists. Finally, we analyze the keywords we find blocked to
682        provide insight into the application-layer blocking ecosystem and
683        compare countries&rsquo; behavior. We find that the most consistently
684        blocked services are related to circumvention tools, pornography, and
685        gambling, but that there is significant country-to-country
686        variation.</p>
687      </div>
688    </div>
689
690    <div class="citation">
691      <div class="title"><a href="/pub/papers/misissuance-sp18.pdf">Tracking Certificate Misissuance in the Wild</a></div>
692      <div class="author">Deepak&nbsp;Kumar, Zhengping&nbsp;Wang, Matthew&nbsp;Hyder, Joseph&nbsp;Dickinson, Gabrielle&nbsp;Beck, David&nbsp;Adrian, Joshua&nbsp;Mason, Zakir&nbsp;Durumeric, J.&nbsp;Alex&nbsp;Halderman, and Michael&nbsp;Bailey</div>
693      <div class="ref"><i>39th IEEE Symposium on Security and Privacy</i> (Oakland &rsquo;18), May 2018</div>
694      <div class="abstract">
695        <p>Certificate Authorities (CAs) regularly make mechanical errors when
696        issuing certificates. To quantify these errors, we introduce ZLint, a
697        certificate linter that codifies the policies set forth by the
698        CA/Browser Forum Baseline Requirements and RFC 5280 that can be tested
699        in isolation. We run ZLint on browser-trusted certificates in Censys and
700        systematically analyze how well CAs construct certificates. We find that
701        the number errors has drastically reduced since 2012. In 2017, only
702        0.02% of certificates have errors. However, this is largely due to a
703        handful of large authorities that consistently issue correct
704        certificates. There remains a long tail of small authorities that
705        regularly issue non-conformant certificates. We further find that
706        issuing certificates with errors is correlated with other types of
707        mismanagement and for large authorities, browser action. Drawing on our
708        analysis, we conclude with a discussion on how the community can best
709        use lint data to identify authorities with worrisome organizational
710        practices and ensure long-term health of the Web PKI.</p>
711      </div>
712    </div>
713
714    <div class="citation">
715      <div class="title"><a href="/pub/papers/snet-imc17.pdf">Initial Measurements of the Cuban Street Network</a></div>
716      <div class="author">Eduardo&nbsp;Pujol, Will&nbsp;Scott, Eric&nbsp;Wustrow, and J.&nbsp;Alex&nbsp;Halderman</div>
717      <div class="ref"><i>17th ACM Internet Measurement Conference</i> (IMC &rsquo;17), November 2017</div>
718      <div class="abstract">
719        <p>Internet access in Cuba is severely constrained, due to limited
720        availability, slow speeds, and high cost. Within this isolated
721        environment, technology enthusiasts have constructed a disconnected but
722        vibrant IP network that has grown organically to reach tens of thousands
723        of households across Havana. We present the first detailed
724        characterization of this deployment, which is known as the SNET, or
725        Street Network. Working in collaboration with SNET operators, we
726        describe the network&rsquo;s infrastructure and map its topology, and we
727        measure bandwidth, available services, usage patterns, and user
728        demographics. Qualitatively, we attempt to answer <i>why</i> the SNET
729        exists and what benefits it has afforded its users. We go on to discuss
730        technical challenges the network faces, including scalability, security,
731        and organizational issues. To our knowledge, the SNET is the largest
732        isolated community-driven network in existence, and its structure,
733        successes, and obstacles show fascinating contrasts and similarities to
734        those of the Internet at large.</p>
735      </div>
736    </div>
737
738    <div class="citation">
739      <div class="title"><a href="https://arxiv.org/abs/1707.08619">Public Evidence from Secret Ballots</a></div>
740      <div class="author">Matthew&nbsp;Bernhard, Josh&nbsp;Benaloh, J.&nbsp;Alex&nbsp;Halderman, Ronald&nbsp;L.&nbsp;Rivest, Peter&nbsp;Y.&nbsp;A.&nbsp;Ryan, Philip&nbsp;B.&nbsp;Stark, Vanessa&nbsp;Teague, Poorvi&nbsp;L.&nbsp;Vora, and Dan&nbsp;S.&nbsp;Wallach</div>
741      <div class="ref"><i>2nd International Joint Conference on Electronic Voting</i> (E-Vote-ID &rsquo;17), October 2017</div>
742      <div class="abstract">
743        <p>Elections seem simple&mdash;<i>aren&rsquo;t they just counting?</i>
744        But they have a unique, challenging combination of security and privacy
745        requirements. The stakes are high; the context is adversarial; the
746        electorate needs to be convinced that the results are correct; and the
747        secrecy of the ballot must be ensured. And they have practical
748        constraints: time is of the essence, and voting systems need to be
749        affordable and maintainable, and usable by voters, election officials,
750        and pollworkers. It is thus not surprising that voting is a rich
751        research area spanning theory, applied cryptography, practical systems
752        analysis, usable security, and statistics. Election integrity involves
753        two key concepts: <i>convincing evidence that outcomes are correct and
754        privacy</i>, which amounts to <i>convincing assurance that there is no
755        evidence</i> about how any given person voted. These are obviously in
756        tension. We examine how current systems walk this tightrope.</p>
757      </div>
758    </div>
759
760    <div class="citation">
761      <div class="title"><a href="/pub/papers/mirai-sec17.pdf">Understanding the Mirai Botnet</a></div>
762      <div class="author">Manos&nbsp;Antonakakis, Tim&nbsp;April, Michael&nbsp;Bailey, Matt&nbsp;Bernhard, Elie&nbsp;Bursztein, Jaime&nbsp;Cochran, Zakir&nbsp;Durumeric, J.&nbsp;Alex&nbsp;Halderman, Luca&nbsp;Invernizzi, Michalis&nbsp;Kallitsis, Deepak&nbsp;Kumar, Chaz&nbsp;Lever, Zane&nbsp;Ma, Joshua&nbsp;Mason, Damian&nbsp;Menscher, Chad&nbsp;Seaman, Nick&nbsp;Sullivan, Kurt&nbsp;Thomas, and Yi&nbsp;Zhou</div>
763      <div class="ref"><i>26th USENIX Security Symposium</i> (Sec &rsquo;17), August 2017</div>
764      <div class="abstract">
765        <p>The Mirai botnet, composed primarily of embedded and IoT devices,
766        took the Internet by storm in late 2016 when it overwhelmed several
767        high-profile targets with massive distributed denial-of-service (DDoS)
768        attacks. In this paper, we provide a seven-month retrospective analysis
769        of Mirai&rsquo;s growth to a peak of 600k infections and a history of
770        its DDoS victims. By combining a variety of measurement perspectives, we
771        analyze how the botnet emerged, what classes of devices were affected,
772        and how Mirai variants evolved and competed for vulnerable hosts. Our
773        measurements serve as a lens into the fragile ecosystem of IoT devices.
774        We argue that Mirai may represent a sea change in the evolutionary
775        development of botnets&mdash;the simplicity through which devices were
776        infected and its precipitous growth, demonstrate that novice malicious
777        techniques can compromise enough low-end devices to threaten even some
778        of the best-defended targets. To address this risk, we recommend
779        technical and nontechnical interventions, as well as propose future
780        research directions.</p>
781      </div>
782    </div>
783
784    <div class="citation">
785      <div class="title"><a href="/pub/papers/deployment-foci17.pdf">An ISP-Scale Deployment of TapDance</a></div>
786      <div class="author">Sergey&nbsp;Frolov, Fred&nbsp;Douglas, Will&nbsp;Scott, Allison&nbsp;McDonald, Benjamin&nbsp;VanderSloot, Rod&nbsp;Hynes, Adam&nbsp;Kruger, Michalis&nbsp;Kallitsis, David&nbsp;Robinson, Nikita&nbsp;Borisov, J.&nbsp;Alex&nbsp;Halderman, and Eric&nbsp;Wustrow</div>
787      <div class="ref"><i>7th USENIX Workshop on Free and Open Communications on the Internet</i> (FOCI &rsquo;17), August 2017</div>
788      <div class="abstract">
789        <p>We report initial results from the world&rsquo;s first ISP-scale
790        field trial of a refraction networking system. Refraction networking is
791        a next-generation censorship circumvention approach that locates proxy
792        functionality in the middle of the network, at participating ISPs or
793        other network operators. We built a high-performance implementation of
794        the TapDance refraction networking scheme and deployed it on four ISP
795        uplinks with an aggregate bandwidth of 100 Gbps. Over one week of
796        operation, our deployment served more than 50,000 real users. The
797        experience demonstrates that TapDance can be practically realized at ISP
798        scale with good performance and at a reasonable cost, potentially paving
799        the way for long-term, large-scale deployments of TapDance or other
800        refraction networking schemes in the future.</p>
801      </div>
802    </div>
803
804    <div class="citation">
805      <div class="title"><a href="/pub/papers/tangled-www17.pdf">Security Challenges in an Increasingly Tangled Web</a></div>
806      <div class="author">Deepak&nbsp;Kumar, Zane&nbsp;Ma, Zakir&nbsp;Durumeric, Ariana&nbsp;Mirian, Joshua&nbsp;Mason, J.&nbsp;Alex&nbsp;Halderman, and Michael&nbsp;Bailey</div>
807      <div class="ref"><i>26th World Wide Web Conference</i> (WWW &rsquo;17), April 2017</div>
808      <div class="abstract">
809        <p>Over the past 20 years, websites have grown increasingly complex and
810        interconnected. In 2016, only a negligible number of sites are
811        dependency free, and over 90% of sites rely on external 
811content. In this
812        paper, we investigate the current state of web dependencies and explore
813        two security challenges associated with the increasing reliance on
814        external services: (1) the expanded attack surface associated with
815        serving unknown, implicitly trusted third-party content, and (2) how the
816        increased set of external dependencies impacts HTTPS adoption. We hope
817        that by shedding light on these issues, we can encourage developers to
818        consider the security risks associated with serving third-party content
819        and prompt service providers to more widely deploy HTTPS.</p>
820      </div>
821    </div>
822
823    <div class="citation">
824      <div class="title"><a href="/pub/papers/interception-ndss17.pdf">The Security Impact of HTTPS Interception</a></div>
825      <div class="author">Zakir&nbsp;Durumeric, Zane&nbsp;Ma, Drew&nbsp;Springall, Richard&nbsp;Barnes, Nick&nbsp;Sullivan, Elie&nbsp;Bursztein, Michael&nbsp;Bailey, J.&nbsp;Alex&nbsp;Halderman, and Vern&nbsp;Paxson</div>
826      <div class="ref"><i>24th Network and Distributed Systems Symposium</i> (NDSS &rsquo;17), February 2017</div>
827      <div class="abstract">
828        <p>As HTTPS deployment grows, middlebox and antivirus products are
829        increasingly intercepting TLS connections to retain visibility into
830        network traffic. In this work, we present a comprehensive study on the
831        prevalence and impact of HTTPS interception. First, we show that web
832        servers can detect interception by identifying a mismatch between the
833        HTTP User-Agent header and TLS client behavior. We characterize the TLS
834        handshakes of major browsers and popular interception products, which we
835        use to build a set of heuristics to detect interception and identify the
836        responsible product. We deploy these heuristics at three large network
837        providers: (1) Mozilla Firefox update servers, (2) a set of popular
838        e-commerce sites, and (3) the Cloudflare content distribution network.
839        We find more than an order of magnitude more interception than
840        previously estimated and with dramatic impact on connection security. To
841        understand why security suffers, we investigate popular middleboxes and
842        client- side security software, finding that nearly all reduce
843        connection security and many introduce severe vulnerabilities. Drawing
844        on our measurements, we conclude with a discussion on recent proposals
845        to safely monitor HTTPS and recommendations for the security
846        community.</p>
847      </div>
848    </div>
849
850    <div class="citation">
851      <div class="title"><a href="/pub/papers/subgroup-ndss16.pdf">Measuring Small Subgroup Attacks Against Diffie-Hellman</a> <span class="altlink">(<a href="https://eprint.iacr.org/2016/995">ePrint</a>)</span></div>
852      <div class="author">Luke&nbsp;Valenta, David&nbsp;Adrian, Antonio&nbsp;Sanso, Shaanan&nbsp;Cohney, Joshua&nbsp;Fried, Marcella&nbsp;Hastings, J.&nbsp;Alex&nbsp;Halderman, and Nadia&nbsp;Heninger</div>
853      <div class="ref"><i>24th Network and Distributed Systems Symposium</i> (NDSS &rsquo;17), February 2017</div>
854      <div class="abstract">
855        <p>Several recent standards, including NIST SP 800-56A and RFC 5114,
856        advocate the use of &ldquo;DSA&rdquo; parameters for Diffie-Hellman key
857        exchange. While it is possible to use such parameters securely,
858        additional validation checks are necessary to prevent well known and
859        potentially devastating attacks. In this paper, we observe that many
860        Diffie-Hellman implementations do not properly validate key exchange
861        inputs. Combined with other protocol properties and implementation
862        choices, this can radically decrease security. We measure the prevalence
863        of these parameter choices in the wild for HTTPS, POP3S, SMTP with
864        STARTTLS, SSH, IKEv1, and IKEv2, finding millions of hosts using DSA and
865        other non-&ldquo;safe&rdquo; primes for Diffie-Hellman key exchange,
866        many of them in combination with potentially vulnerable behaviors. We
867        examine over 20 open-source cryptographic libraries and applications and
868        observe that until January 2016, not a single one validated subgroup
869        orders by default. We found feasible full or partial key recovery
870        vulnerabilities in OpenSSL, the Exim mail server, the Unbound DNS
871        client, and Amazon&rsquo;s load balancer, as well as susceptibility to
872        weaker attacks in many other applications.</p>
873      </div>
874    </div>
875
876    <div class="citation">
877      <div class="title"><a href="/pub/papers/ics-pst16.pdf">An Internet-Wide View of ICS Devices</a></div>
878      <div class="author">Ariana&nbsp;Mirian, Zane&nbsp;Ma, David&nbsp;Adrian, Matthew&nbsp;Tischer, Thasphon&nbsp;Chuenchujit, Tim&nbsp;Yardley, Robin&nbsp;Berthier, Josh&nbsp;Mason, Zakir&nbsp;Durumeric, J.&nbsp;Alex&nbsp;Halderman, and Michael&nbsp;Bailey</div>
879      <div class="ref"><i>14th IEEE Conference on Privacy, Security, and Trust</i> (PST &rsquo;16), December 2016</div>
880      <div class="abstract">
881        <p>
881Industrial control systems have become ubiquitous, enabling the
882        remote, electronic control of physical equipment and sensors. Originally
883        designed to operate on closed networks, the protocols used by these
884        devices have no built-in security. However, despite this, an alarming
885        number of systems are connected to the public Internet and an attacker
886        who finds a device often can cause catastrophic damage to physical
887        infrastructure. We consider two aspects of ICS security in this work:
888        (1) what devices have been inadvertently exposed on the public Internet,
889        and (2) who is searching for vulnerable systems. First, we implement
890        five common SCADA protocols in ZMap and conduct a survey of the public
891        IPv4 address space finding more than 60K publicly accessible systems.
892        Second, we use a large network telescope and high-interaction honeypots
893        to find and profile actors searching for devices. We hope that our
894        findings can both motivate and inform future work on securing industrial
895        control systems.</p>
896      </div>
897    </div>
898
899    <div class="citation">
900      <div class="title"><a href="/pub/papers/kiosk-pst16.pdf">Implementing Attestable Kiosks</a></div>
901      <div class="author">Matthew&nbsp;Bernhard, J.&nbsp;Alex&nbsp;Halderman, and Gabe&nbsp;Stocco</div>
902      <div class="ref"><i>14th IEEE Conference on Privacy, Security, and Trust</i> (PST &rsquo;16), December 2016</div>
903      <div class="abstract">
904        <p>In this paper we explore the notion of a secure kiosk, a trusted
905        computing platform built using off-the-shelf components. We demonstrate
906        how kiosks serve as convenient primitives when designing secure
907        computing protocols, as they allow for a very prescribed set of
908        assumptions to be made about a system. We begin by defining the
909        necessary properties of a kiosk, and then explain how each of these
910        properties can (or cannot) be attained using current off-the-shelf
911        hardware and software components. We construct a proof-of-concept
912        implementation using TPM hardware and Windows 10. We also provide
913        ASKVote, the Attestable and Secure Voting protocol to demonstrate the
914        flexibility gained from the use of kiosks in a larger secure system.</p>
915      </div>
916    </div>
917
918    <div class="citation">
919      <div class="title"><a href="/pub/papers/police-pst16.pdf">A Security Analysis of Police Computer Systems</a></div>
920      <div class="author">Benjamin&nbsp;VanderSloot, Stuart&nbsp;Wheaton, and J.&nbsp;Alex&nbsp;Halderman</div>
921      <div class="ref"><i>14th IEEE Conference on Privacy, Security, and Trust</i> (PST &rsquo;16), December 2016</div>
922      <div class="abstract">
923        <p></p>
924      </div>
925    </div>
926
927    <div class="citation">
928      <div class="title"><a href="/pub/papers/forward-secrecy-imc16.pdf">Measuring the Security Harm of TLS Crypto Shortcuts</a></div>
929      <div class="author">Drew&nbsp;Springall, Zakir&nbsp;Durumeric, and J.&nbsp;Alex&nbsp;Halderman</div>
930      <div class="ref"><i>16th ACM Internet Measurement Conference</i> (IMC &rsquo;16), November 2016</div>
931      <div class="abstract">
932        <p>TLS has the potential to provide strong protection against
933        network-based attackers and mass surveillance, but many implementations
934        take security shortcuts in order to reduce the costs of cryptographic
935        computations and network round trips. We report the results of a
936        nine-week study that measures the use and security impact of these
937        shortcuts for HTTPS sites among Alexa Top Million domains. We find
938        widespread deployment of DHE and ECDHE private value reuse, TLS session
939        resumption, and TLS session tickets. These practices greatly reduce the
940        protection afforded by forward secrecy: connections to 38% of Top
941        Million HTTPS sites are vulnerable to decryption if the server is
942        compromised up to 24 hours later, and 10% up to 30 days later,
943        regardless of the selected cipher suite. We also investigate the
944        practice of TLS secrets and session state being shared across domains,
945        finding that in some cases, the theft of a single secret value can
946        compromise connections to tens of thousands of sites. These results
947        suggest that site operators need to better understand the tradeoffs
948        between optimizing TLS performance and providing strong security,
949        particularly when faced with nation-state attackers with a history of
950        aggressive, large-scale surveillance.</p>
951      </div>
952    </div>
953
954    <div class="citation">
955      <div class="title"><a href="/pub/papers/https-perspectives-imc16.pdf">Towards a Complete View of the Certificate Ecosystem</a></div>
956      <div class="author">Benjamin&nbsp;VanderSloot, Johanna&nbsp;Amann, Matthew&nbsp;Bernhard, Zakir&nbsp;Durumeric, Michael&nbsp;Bailey, and J.&nbsp;Alex&nbsp;Halderman</div>
957      <div class="ref"><i>16th ACM Internet Measurement Conference</i> (IMC &rsquo;16), November 2016</div>
958      <div class="abstract">
959        <p>The HTTPS certificate ecosystem has been of great interest to the
960        measurement and security communities. Without any ground truth,
961        researchers have attempted to study this PKI from a variety of
962        fragmented perspectives, including passively monitored networks, scans
963        of the popular domains or the IPv4 address space, search engines such as
964        Censys, and Certificate Transparency (CT) logs. In this work, we
965        comparatively analyze all these perspectives. We find that aggregated CT
966        logs and Censys snapshots have many properties that complement each
967        other, and that together they encompass over 99% of all certificates
968        found by any of these techniques. However, they still miss 1.5% of
969        certificates observed in a crawl of all domains in <span
970        class="tt">.com</span>, <span class="tt">.net</span>, and <span
971        class="tt">.org</span>. We go on to illustrate how this combined
972        perspective affects results from previous studies. In light of these
973        findings, we have worked with the operators of Censys to incorporate CT
974        log data into its results going forward, and we recommend that future
975        HTTPS measurement adopt this new vantage.</p>
976      </div>
977    </div>
978
979    <div class="citation">
980      <div class="title"><a href="/pub/papers/cbsec-nspw16.pdf">Content-Based Security for the Web</a></div>
981      <div class="author">Alexander&nbsp;Afanasyev, J.&nbsp;Alex&nbsp;Halderman, Scott&nbsp;Ruoti, Kent&nbsp;Seamons, Yingdi&nbsp;Yu, Daniel&nbsp;Zappala, and Lixia&nbsp;Zhang</div>
982      <div class="ref"><i>2016 New Security Paradigms Workshop</i> (NSPW &rsquo;16), September 2016</div>
983      <div class="abstract">
984        <p>The World Wide Web has become the most common platform for building
985        applications and delivering content. Yet despite years of research, the
986        web continues to face severe security challenges related to data
987        integrity and confidentiality. Rather than continuing the
988        exploit-and-patch cycle, we propose addressing these challenges at an
989        architectural level, by supplementing the web&rsquo;s existing
990        connection-based and server-based security models with a new approach:
991        content-based security. With this approach, content is directly signed
992        and encrypted at rest, enabling it to be delivered via any path and then
993        validated by the browser. We explore how this new architectural approach
994        can be applied to the web and analyze its security benefits. We then
995        discuss a broad research agenda to realize this vision and the
996        challenges that must be overcome.</p>
997      </div>
998    </div>
999
1000    <div class="citation">
1001      <div class="title"><a href="https://drownattack.com/drown-attack-paper.pdf">DROWN: Breaking TLS using SSLv2</a> <span class="altlink">(<a href="https://drownattack.com/">website and test tool</a>)</span></div>
1002      <div class="author">Nimrod&nbsp;Aviram, Sebastian&nbsp;Schinzel, Juraj&nbsp;Somorovsky, Nadia&nbsp;Heninger, Maik&nbsp;Dankel, Jens&nbsp;Steube, Luke&nbsp;Valenta, David&nbsp;Adrian, J.&nbsp;Alex&nbsp;Halderman, Viktor&nbsp;Dukhovni, Emilia&nbsp;K&auml;sper, Shaanan&nbsp;Cohney, Susanne&nbsp;Engels, Christof&nbsp;Paar, and Yuval&nbsp;Shavitt</div>
1003      <div class="ref"><i>25th USENIX Security Symposium</i> (Sec &rsquo;16), Austin, TX, August 2016 &mdash; <b><a href="https://pwnies.com/">Pwnie&nbsp;Award&nbsp;</a></b> &mdash; <b>Internet Defense Prize Finalist</b></div>
1004      <div class="abstract">
1005        <p>We present DROWN, a novel cross-protocol attack on TLS that uses a
1006        server supporting SSLv2 as an oracle to decrypt modern TLS
1007        connections.</p>
1008        <p>We introduce two versions of the attack. The more general form
1009        exploits multiple unnoticed protocol flaws in SSLv2 to develop a new and
1010        stronger variant of the Bleichenbacher RSA padding-oracle attack. To
1011        decrypt a 2048-bit RSA TLS ciphertext, an attacker must observe 1,000
1012        TLS handshakes, initiate 40,000 SSLv2 connections, and perform
1013        2<sup>50</sup> offline work. The victim client never initiates SSLv2
1014        connections. We implemented the attack and can decrypt a TLS 1.2
1015        handshake using 2048-bit RSA in under 8 hours, at a cost of $440 on
1016        Amazon EC2. Using Internet-wide scans, we find that 33% of all HTTPS
1017        servers and 22% of those with browser-trusted certificates are
1018        vulnerable to this protocol-level attack due to widespread key and
1019        certificate reuse.</p>
1020        <p>For an even cheaper attack, we apply our new techniques together with
1021        a newly discovered vulnerability in OpenSSL that was present in releases
1022        from 1998 to early 2015. Given an unpatched SSLv2 server to use as an
1023        oracle, we can decrypt a TLS ciphertext in one minute on a single
1024        CPU&mdash;fast enough to enable man-in-the-middle attacks against modern
1025        browsers. We find that 26% of HTTPS servers are vulnerable to this
1026        attack.</p>
1027        <p>We further observe that the QUIC protocol is vulnerable to a variant
1028        of our attack that allows an attacker to impersonate a server
1029        indefinitely after performing as few as 2<sup>17</sup> SSLv2 connections
1030        and 2<sup>58</sup> offline work.</p>
1031        <p>We conclude that SSLv2 is not only weak, but actively harmful to the
1032        TLS ecosystem.</p>
1033      </div>
1034    </div>
1035
1036    <div class="citation">
1037      <div class="title"><a href="/pub/papers/ftp-dsn16.pdf">FTP: The Forgotten Cloud</a></div>
1038      <div class="author">Drew&nbsp;Springall, Zakir&nbsp;Durumeric, and J.&nbsp;Alex&nbsp;Halderman</div>
1039      <div class="ref"><i>46th IEEE/IFIP International Conference on Dependable Systems and Networks</i> (DSN &rsquo;16), Toulouse, June 2016</div>
1040      <div class="abstract">
1041        <p>Once pervasive, the File Transfer Protocol (FTP) has been largely
1042        supplanted by HTTP, SCP, and BitTorrent for transferring data between
1043        hosts. Yet, in a comprehensive analysis of the FTP ecosystem as of 2015,
1044        we find that there are still more than 13 million FTP servers in the
1045        IPv4 address space, 1.1 million of which allow &ldquo;anonymous&rdquo;
1046        (public) access. These anonymous FTP servers leak sensitive information,
1047        such as tax documents and cryptographic secrets. More than 20,000 FTP
1048        servers allow public write access, which has facilitated malicious
1049        actors&rsquo; use of free storage as well as malware deployment and
1050        click-fraud attacks. We further investigate real-world attacks by
1051        deploying eight FTP honeypots, shedding light on how attackers are
1052        abusing and exploiting vulnerable servers. We conclude with lessons and
1053        recommendations for securing FTP.</p>
1054      </div>
1055    </div>
1056
1057    <div class="citation">
1058      <div class="title"><a href="/pub/papers/deception-fc16.pdf">Android UI Deception Revisited: Attacks and Defenses</a></div>
1059      <div class="author">Earlence&nbsp;Fernandes, Qi&nbsp;Alfred&nbsp;Chen, Justin&nbsp;Paupore, Georg&nbsp;Essl, J.&nbsp;Alex&nbsp;Halderman, Z.&nbsp;Morley&nbsp;Mao, and Atul&nbsp;Prakash</div>
1060      <div class="ref"><i>20th International Conference on Financial Cryptography and Data Security</i> (FC &rsquo;16), Barbados, February 2016</div>
1061      <div class="abstract">
1062        <p>App-based deception attacks are increasingly a problem on mobile
1063        devices and they are used to steal passwords, credit card numbers, text
1064        messages, etc. Current versions of Android are susceptible to these
1065        attacks. Recently, Bianchi et al. proposed a novel solution (&ldquo;What
1066        the App is That&rdquo;) that included a host-based system to identify
1067        apps to users via a security indicator and help assure them that their
1068        input goes to the identified apps. Unfortunately, we found that the
1069        solution has a significant side channel vulnerability as well as
1070        susceptibility to clickjacking that allow non-privileged malware to
1071        completely compromise the defenses, and successfully steal passwords or
1072        other keyboard input. We discuss the vulnerabilities found, propose
1073        possible defenses, and then evaluate the defenses against different
1074        types of UI deception attacks.</p>
1075      </div>
1076    </div>
1077
1078    <div class="citation">
1079      <div class="title"><a href="https://weakdh.org/imperfect-forward-secrecy.pdf">Imperfect Forward Secrecy: How Diffie-Hellman Fails in Practice</a> <span class="altlink">(<a href="https://weakdh.org/">website</a>)</span></div>
1080      <div class="author">David&nbsp;Adrian, Karthikeyan&nbsp;Bhargavan, Zakir&nbsp;Durumeric, Pierrick&nbsp;Gaudry, Matthew&nbsp;Green, J.&nbsp;Alex&nbsp;Halderman, Nadia&nbsp;Heninger, Drew&nbsp;Springall, Emmanuel&nbsp;Thom&eacute;, Luke&nbsp;Valenta, Benjamin&nbsp;VanderSloot, Eric&nbsp;Wustrow, Santiago&nbsp;Zanella-B&eacute;guelin, and Paul&nbsp;Zimmermann</div>
1081      <div class="ref"><i>22nd ACM Conference on Computer and Communications Security</i> (CCS &rsquo;15), Denver, CO, October 2015 &mdash; <b>Best&nbsp;Paper&nbsp;Award</b> &mdash; <b>
1081<a href="https://pwnies.com/">Pwnie&nbsp;Award&nbsp;</a></b></div>
1082      <div class="ref">Reprinted in <i><a href="/pub/papers/weakdh-cacm19.pdf">Communications of the ACM</a></i>, <!--52(5):91-98,--> January 2019</div>
1083      <div class="abstract">
1084        <p>We investigate the security of Diffie-Hellman key exchange as used in
1085        popular Internet protocols and find it to be less secure than widely
1086        believed. First, we present Logjam, a novel flaw in TLS that lets a
1087        man-in-the-middle downgrade connections to &ldquo;export-grade&rdquo;
1088        Diffie-Hellman. To carry out this attack, we implement the number field
1089        sieve discrete log algorithm. After a week-long precomputation for a
1090        specified 512-bit group, we can compute arbitrary discrete logs in that
1091        group in about a minute. We find that 82% of vulnerable servers use a
1092        single 512-bit group, allowing us to compromise connections to 7% of
1093        Alexa Top Million HTTPS sites. In response, major browsers are being
1094        changed to reject short groups.</p>
1095        <p>We go on to consider Diffie-Hellman with 768- and 1024-bit groups. We
1096        estimate that even in the 1024-bit case, the computations are plausible
1097        given nation-state resources. A small number of fixed or standardized
1098        groups are used by millions of servers; performing precomputation for a
1099        single 1024-bit group would allow passive eavesdropping on 18% of
1100        popular HTTPS sites, and a second group would allow decryption of
1101        traffic to 66% of IPsec VPNs and 26% of SSH servers. A close reading of
1102        published NSA leaks shows that the agency&rsquo;s attacks on VPNs are
1103        consistent with having achieved such a break. We conclude that moving to
1104        stronger key exchange methods should be a priority for the Internet
1105        community.</p>
1106      </div>
1107    </div>
1108
1109    <div class="citation">
1110      <div class="title"><a href="/pub/papers/censys-ccs15.pdf">Censys: A Search Engine Backed by Internet-Wide Scanning</a> <span class="altlink">(<a href="https://censys.io">website</a>)</span></div>
1111      <div class="author">Zakir&nbsp;Durumeric, David&nbsp;Adrian, Ariana&nbsp;Mirian, Michael&nbsp;Bailey, and J.&nbsp;Alex&nbsp;Halderman</div>
1112      <div class="ref"><i>22nd ACM Conference on Computer and Communications Security</i> (CCS &rsquo;15), Denver, CO, October 2015</div>
1113      <div class="abstract">
1114        <p>Fast Internet-wide scanning has opened new avenues for security
1115        research, ranging from uncovering widespread vulnerabilities in random
1116        number generators to tracking the evolving impact of Heartbleed.
1117        However, this technique still requires significant effort: even simple
1118        questions, such as, &ldquo;What models of embedded devices prefer CBC
1119        ciphers?&rdquo;, require developing an application scanner, manually
1120        identifying and tagging devices, negotiating with network
1121        administrators, and responding to abuse complaints. In this paper, we
1122        introduce Censys, a public search engine and data processing facility
1123        backed by data collected from ongoing Internet-wide scans. Designed to
1124        help researchers answer security-related questions, Censys supports
1125        full-text searches on protocol banners and querying a wide range of
1126        derived fields (e.g., <span style="font-family:monospace; font-size:80%;">443.https.cipher</span>). It can identify specific
1127        vulnerable devices and networks and generate statistical reports on
1128        broad usage patterns and trends. Censys returns these results in
1129        sub-second time, dramatically reducing the effort of understanding the
1130        hosts that comprise the Internet. We present the search engine
1131        architecture and experimentally evaluate its performance. We also
1132        explore Censys&rsquo;s applications and show how questions asked in
1133        recent studies become simple to answer.</p>
1134      </div>
1135    </div>
1136
1137    <div class="citation">
1138      <div class="title"><a href="/pub/papers/mail-imc15.pdf">Neither Snow Nor Rain Nor MITM&hellip; An Empirical Analysis of Email Delivery Security</a></div>
1139      <div class="author">Zakir&nbsp;Durumeric, David&nbsp;Adrian, Ariana&nbsp;Mirian, James&nbsp;Kasten, Elie&nbsp;Bursztein, Nicholas&nbsp;Lidzborski, Kurt&nbsp;Thomas, Vijay&nbsp;Eranti, Michael&nbsp;Bailey, and J.&nbsp;Alex&nbsp;Halderman</div>
1140      <div class="ref"><i>15th ACM Internet Measurement Conference</i> (IMC &rsquo;15), Tokyo, Japan, October 2015 &mdash; <b><a href="https://irtf.org/anrp">IRTF Applied Networking Research Prize</a> Winner</b></div>
1141      <div class="abstract">
1142        <p>The SMTP protocol is responsible for carrying some of users&rsquo;
1143        most intimate communication, but like other Internet protocols,
1144        authentication and confidentiality were added only as an afterthought.
1145        In this work, we present the first report on global adoption rates of
1146        SMTP security extensions, including StartTLS, SPF, DKIM, and DMARC. We
1147        present data from two perspectives: SMTP server configurations for the
1148        Alexa Top Million domains, and over a year of SMTP connections to and
1149        from Gmail. We find that the top mail providers (e.g., Gmail, Yahoo,
1150        Outlook) all proactively encrypt and authenticate messages. However,
1151        these best practices have yet to reach widespread adoption in a long
1152        tail of over 700,000 SMTP servers, of which only 35% successfully
1153        configure encryption and 1.1% specify a DMARC authentication policy.
1154        This security patchwork&mdash;paired with SMTP policies that favor
1155        failing open to allow gradual deployment&mdash;
1155exposes users to
1156        attackers who downgrade TLS connections in favor of cleartext and who
1157        falsify MX records to reroute messages. We present evidence of such
1158        attacks in the wild, highlighting seven countries where more than 20% of
1159        inbound Gmail messages arrive in cleartext due to network attackers.</p>
1160      </div>
1161    </div>
1162
1163    <div class="citation">
1164      <div class="title"><a href="https://arxiv.org/pdf/1504.05646v2.pdf">The New South Wales iVote System: Security Failures and Verification Flaws in a Live Online Election</a> <span class="altlink">(<a href="https://freedom-to-tinker.com/blog/teaguehalderman/ivote-vulnerability/">blog post</a>)</span></div>
1165      <div class="author">J.&nbsp;Alex&nbsp;Halderman and Vanessa&nbsp;Teague</div>
1166      <div class="ref"><i>5th International Conference on E-voting and Identity</i> (VoteID &rsquo;15), Bern, Switzerland, September 2015</div>
1167      <div class="abstract">
1168        <p>In the world&rsquo;s largest-ever deployment of online voting, the
1169        iVote Internet voting system was trusted for the return of 280,000
1170        ballots in the 2015 state election in New South Wales, Australia. During
1171        the election, we performed an independent security analysis of parts of
1172        the live iVote system and uncovered severe vulnerabilities that could be
1173        leveraged to manipulate votes, violate ballot privacy, and subvert the
1174        verification mechanism. These vulnerabilities do not seem to have been
1175        detected by the election authorities before we disclosed them, despite a
1176        pre-election security review and despite the system having run in a live
1177        state election for five days. One vulnerability, the result of including
1178        analytics software from an insecure external server, exposed some votes
1179        to complete compromise of privacy and integrity. At least one
1180        parliamentary seat was decided by a margin much smaller than the number
1181        of votes taken while the system was vulnerable. We also found protocol
1182        flaws, including vote verification that was itself susceptible to
1183        manipulation. This incident underscores the difficulty of conducting
1184        secure elections online and carries lessons for voters, election
1185        officials, and the e-voting research community.</p>
1186      </div>
1187    </div>
1188
1189    <div class="citation">
1190      <div class="title"><a href="/pub/papers/umbra-woscps15.pdf">Umbra: Embedded Web Security through Application-Layer Firewalls</a></div>
1191      <div class="author">Travis&nbsp;Finkenauer and J.&nbsp;Alex&nbsp;Halderman</div>
1192      <div class="ref"><i>1st Workshop on the Security of Cyberphysical Systems</i> (WOS-CPS &rsquo;15), Vienna, Austria, September 2015</div>
1193      <div class="abstract">
1194        <p>Embedded devices with web interfaces are prevalent, but, due to
1195        memory and processing constraints, implementations typically make use of
1196        Common Gateway Interface (CGI) binaries written in low-level,
1197        memory-unsafe languages. This creates the possibility of memory
1198        corruption attacks as well as traditional web attacks. We present Umbra,
1199        an application-layer firewall specifically designed for protecting web
1200        interfaces in embedded devices. By acting as a &ldquo;friendly
1201        man-in-the-middle,&rdquo; Umbra can protect against attacks such as
1202        cross-site request forgery (CSRF), information leaks, and authentication
1203        bypass vulnerabilities. We evaluate Umbra&rsquo;s security by analyzing
1204        recent vulnerabilities listed in the CVE database from several embedded
1205        device vendors and find that it would have prevented half of these
1206        vulnerabilities. We also show that Umbra comfortably runs within the
1207        constraints of an embedded system while incurring minimal performance
1208        overhead.</p>
1209      </div>
1210    </div>
1211
1212    <div class="citation">
1213      <div class="title"><a href="/pub/papers/replication-woot15.pdf">Replication Prohibited: Attacking Restricted Keyways with 3D Printing</a></div>
1214      <div class="author">Ben&nbsp;Burgess, Eric&nbsp;Wustrow, and J.&nbsp;Alex&nbsp;Halderman</div>
1215      <div class="ref"><i>9th USENIX Workshop on Offensive Technologies</i> (WOOT &rsquo;15), Washington, DC, August 2015</div>
1216      <div class="abstract">
1217        <p>Several attacks against physical pin-tumbler locks require access to
1218        one or more key blanks to perform. These attacks include bumping,
1219        impressioning, rights amplification, and teleduplication. To mitigate
1220        these attacks, many lock systems rely on restricted keyways and use
1221        blanks that are not sold to the general public, making it harder for
1222        attackers to obtain them. Often the key blank designs themselves are
1223        patented, further discouraging distribution or manufacture by even
1224        skilled machinists. In this paper, we investigate the impact that
1225        emerging rapid-prototyping&mdash;or 3D-printing&mdash;tools have on the
1226        security of these restricted keyway systems. We find that commodity 3D
1227        printers are able to produce key blanks and pre-cut keys with enough
1228        resolution to work in several commonly used pin-tumbler locks and that
1229        their material is strong enough to withstand the requirements to perform
1230        the aforementioned attacks. In addition, in order to demonstrate the low
1231        skill requirements necessary to perform these attacks, we develop a tool
1232        that automatically generates a 3D-printable CAD model of a key blank
1233        using only a single picture of a lock&rsquo;s keyway. This tool allows
1234        us to rapidly manufacture key blanks for restricted keyways that were
1235        previously difficult to make or buy. Finally, we discuss possible
1236        mitigations for these attacks that lock manufacturers, installer
1236s, and
1237        users can perform to protect their assets.</p>
1238      </div>
1239    </div>
1240
1241    <div class="citation">
1242      <div class="title"><a href="/pub/papers/heartbleed-imc14.pdf">The Matter of Heartbleed</a></div>
1243      <div class="author">Zakir&nbsp;Durumeric, Frank&nbsp;Li, James&nbsp;Kasten, Johanna&nbsp;Amann, Jethro&nbsp;Beekman, Mathias&nbsp;Payer, Nicolas&nbsp;Weaver, David&nbsp;Adrian, Vern&nbsp;Paxson, Michael&nbsp;Bailey, and J.&nbsp;Alex&nbsp;Halderman</div>
1244      <div class="ref"><i>14th ACM Internet Measurement Conference</i> (IMC &rsquo;14), Vancouver, BC, November 2014 &mdash; <b>Best&nbsp;Paper&nbsp;Award</b></div>
1245      <div class="abstract">
1246        <p>The Heartbleed vulnerability took the Internet by surprise in April
1247        2014. The vulnerability, one of the most consequential since the advent
1248        of the commercial Internet, allowed attackers to remotely read protected
1249        memory from an estimated 24&ndash;55% of popular HTTPS sites. In this
1250        work, we perform a comprehensive, measurement-based analysis of the
1251        vulnerability&rsquo;s impact, including (1) tracking the vulnerable
1252        population, (2) monitoring patching behavior over time, (3) assessing
1253        the impact on the HTTPS certificate ecosystem, and (4) exposing real
1254        attacks that attempted to exploit the bug. Furthermore, we conduct a
1255        large-scale vulnerability notification experiment involving 150,000
1256        hosts and observe a nearly 50% increase in patching by notified hosts.
1257        Drawing upon these analyses, we discuss what went well and what went
1258        poorly, in an effort to understand how the technical community can
1259        respond more effectively to such events in the future.</p>
1260      </div>
1261    </div>
1262
1263    <div class="citation">
1264      <div class="title"><a href="/pub/papers/ivoting-ccs14.pdf">Security Analysis of the Estonian Internet Voting System</a> <span class="altlink">(<a href="https://estoniaevoting.org">website and videos</a>)</span></div>
1265      <div class="author">Drew&nbsp;Springall, Travis&nbsp;Finkenauer, Zakir&nbsp;Durumeric, Jason&nbsp;Kitcat, Harri&nbsp;Hursti, Margaret&nbsp;MacAlpine, and J.&nbsp;Alex&nbsp;Halderman</div>
1266      <div class="ref"><i>21st ACM Conference on Computer and Communications Security</i> (CCS &rsquo;14), Scottsdale, AZ, November 2014</div>
1267      <div class="abstract">
1268        <p>Estonia was the first country in the world to use Internet voting
1269        nationally, and today more than 30% of its ballots are cast online. In
1270        this paper, we analyze the security of the Estonian I-voting system
1271        based on a combination of in-person election observation, code review,
1272        and adversarial testing. Adopting a threat model that considers the
1273        advanced threats faced by a national election system&mdash;including
1274        dishonest insiders and state-sponsored attacks&mdash;we find that the
1275        I-voting system has serious architectural limitations and procedural
1276        gaps that potentially jeopardize the integrity of elections. In
1277        experimental attacks on a reproduction of the system, we demonstrate how
1278        such attackers could target the election servers or voters&rsquo;
1279        clients to alter election results or undermine the legitimacy of the
1280        system. Our findings illustrate the practical obstacles to Internet
1281        voting in the modern world, and they carry lessons for Estonia, for
1282        other countries considering adopting such systems, and for the security
1283        research community.</p>
1284      </div>
1285    </div>
1286
1287    <div class="citation">
1288      <div class="title"><a href="/pub/papers/audit-evote14.pdf">Efficiently Auditing Multi-Level Elections</a></div>
1289      <div class="author">Joshua&nbsp;A.&nbsp;Kroll, Edward&nbsp;W.&nbsp;Felten, and J.&nbsp;Alex&nbsp;Halderman</div>
1290      <div class="ref"><i>6th International Conference on Electronic Voting</i> (EVOTE &rsquo;14), Lochau, Austria, October 2014</div>
1291      <div class="abstract">
1292        <p>In a multi-level election, voters are divided into groups, an
1293        election is held within each group, and some deterministic procedure is
1294        used to combine the group results to determine the overall election
1295        result. Examples of multi-level elections include U.S. presidential
1296        elections and some parliamentary elections (such as those with regional
1297        groupings of voters). The results of such an election can hinge on a few
1298        votes in one group, while being insensitive to large shifts within other
1299        groups. These disparities create opportunities to focus election
1300        integrity efforts in the places where they have the highest leverage. We
1301        consider how to improve the efficiency of post-election audits, such as
1302        those that compare paper ballots to corresponding electronic records, in
1303        multi-level elections. We evaluate our proposed solutions using data
1304        from past elections.</p>
1305      </div>
1306    </div>
1307
1308    <div class="citation">
1309      <div class="title"><a href="https://radsec.org/secure1000-sec14.pdf">Security Analysis of a Full-Body Scanner</a> <span class="altlink">(<a href="https://radsec.org">website</a>)</span></div>
1310      <div class="author">Keaton&nbsp;Mowery, Eric&nbsp;Wustrow, Tom&nbsp;Wypych, Corey&nbsp;Singleton, Chris&nbsp;Comfort, Eric&nbsp;Rescorla, Stephen&nbsp;Checkoway, J.&nbsp;Alex&nbsp;Halderman, and Hovav&nbsp;Shacham</div>
1311      <div class="ref"><i>23rd USENIX Security Symposium</i> (Sec &rsquo;14), San Diego, CA, August 2014</div>
1312      <div class="abstract">
1313        <p>Advanced imaging technologies are a new class of people screening
1314        systems used at airports and other sensitive environments to detect
1315        metallic as well as nonmetallic contraband. We present the first
1316        independent security evaluation of such a system, the Rapiscan Secure
1317        1000 full-body scanner, which was widely deployed at airport checkpoints
1318        in the U.S. from 2009 until 2013. We find that the system provides weak
1319        protection against adaptive adversaries: It is possible to conceal
1320        knives, guns, and explosives from detection by exploiting properties of
1321        the device&rsquo;s backscatter X-ray technology. We also investigate
1322        cyberphysical threats and propose novel attacks that use malicious
1323        software and hardware to compromise the the effectiveness, safety, and
1324        privacy of the device. Overall, our findings paint a mixed picture of
1325        the Secure 1000 that carries lessons for the design, evaluation, and
1326        operation of advanced imaging technologies, for the ongoing public
1327        debate concerning their use, and for cyberphysical security more
1328        broadly.</p>
1329      </div>
1330    </div>
1331
1332    <div class="citation">
1333      <div class="title"><a href="/pub/papers/tapdance-sec14.pdf">TapDance: End-to-Middle Anticensorship without Flow Blocking</a></div>
1334      <div class="author">Eric&nbsp;Wustrow, Colleen&nbsp;M.&nbsp;Swanson, and J.&nbsp;Alex&nbsp;Halderman</div>
1335      <div class="ref"><i>23rd USENIX Security Symposium</i> (Sec &rsquo;14), San Diego, CA, August 2014</div>
1336      <div class="abstract">
1337        <p>In response to increasingly sophisticated state-sponsored Internet
1338        censorship, recent work has proposed a new approach to censorship
1339        resistance: end-to-middle proxying. This concept, developed in systems
1340        such as Telex, Decoy Routing, and Cirripede, moves anticensorship
1341        technology into the core of the network, at large ISPs outside the
1342        censoring country. In this paper, we focus on two technical obstacles to
1343        the deployment of certain end-to-middle schemes: the need to selectively
1344        block flows and the need to observe both directions of a connection. We
1345        propose a new construction, TapDance, that removes these requirements.
1346        TapDance employs a novel TCP-level technique that allows the
1347        anticensorship station at an ISP to function as a passive network tap,
1348        without an inline blocking component. We also apply a novel
1349        steganographic encoding to embed control messages in TLS ciphertext,
1350        allowing us to operate on HTTPS connections even under asymmetric
1351        routing. We implement and evaluate a TapDance prototype that
1352        demonstrates how the system could function with minimal impact on an
1353        ISP&rsquo;s network operations.</p>
1354      </div>
1355    </div>
1356
1357    <div class="citation">
1358      <div class="title"><a href="/pub/papers/scanning-sec14.pdf">An Internet-Wide View of Internet-Wide Scanning</a></div>
1359      <div class="author">Zakir&nbsp;Durumeric, Michael&nbsp;Bailey, and J.&nbsp;Alex&nbsp;Halderman</div>
1360      <div class="ref"><i>23rd USENIX Security Symposium</i> (Sec &rsquo;14), San Diego, CA, August 2014</div>
1361      <div class="abstract">
1362        <p>
1362While it is widely known that port scanning is widespread, neither
1363        the scanning landscape nor the defensive reactions of network operators
1364        have been measured at Internet scale. In this work, we analyze data from
1365        a large network telescope to study scanning activity from the past year,
1366        uncovering large horizontal scan operations and identifying broad
1367        patterns in scanning behavior. We present an analysis of who is
1368        scanning, what services are being targeted, and the impact of new
1369        scanners on the overall landscape. We also analyze the scanning behavior
1370        triggered by recent vulnerabilities in Linksys routers, OpenSSL, and
1371        NTP. We empirically analyze the defensive behaviors that organizations
1372        employ against scanning, shedding light on who detects scanning
1373        behavior, which networks blacklist scanning, and how scan recipients
1374        respond to scans conducted by researchers. We conclude with
1375        recommendations for institutions performing scans and with implications
1376        of recent changes in scanning behavior for researchers and network
1377        operators.</p>
1378      </div>
1379    </div>
1380
1381    <div class="citation">
1382      <div class="title"><a href="/pub/papers/traffic-woot14.pdf">Green Lights Forever: Analyzing the Security of Traffic Infrastructure</a></div>
1383      <div class="author">William&nbsp;Beyer, Branden&nbsp;Ghena, Allen&nbsp;Hillaker, Jonathan&nbsp;Pevarnek, and J.&nbsp;Alex&nbsp;Halderman</div>
1384      <div class="ref"><i>8th USENIX Workshop on Offensive Technologies</i> (WOOT &rsquo;14), San Diego, CA, August 2014</div>
1385      <div class="abstract">
1386        <p>The safety critical nature of traffic infrastructure requires that it
1387        be secure against computer-based attacks, but this is not always the
1388        case. We investigate a networked traffic signal system currently
1389        deployed in the United States and discover a number of security flaws
1390        that exist due to systemic failures by the designers. We leverage these
1391        flaws to create attacks which gain control of the system, and we
1392        successfully demonstrate them on the deployment in coordination with
1393        authorities. Our attacks show that an adversary can control traffic
1394        infrastructure to cause disruption, degrade safety, or gain an unfair
1395        advantage. We make recommendations on how to improve existing systems
1396        and discuss the lessons learned for embedded systems security in
1397        general.</p>
1398      </div>
1399    </div>
1400
1401    <div class="citation">
1402      <div class="title"><a href="/pub/papers/zmap10gig-woot14.pdf">Zippier ZMap: Internet-Wide Scanning at 10Gbps</a></div>
1403      <div class="author">David&nbsp;Adrian, Zakir&nbsp;Durumeric, Gulshan&nbsp;Singh, and J.&nbsp;Alex&nbsp;Halderman</div>
1404      <div class="ref"><i>8th USENIX Workshop on Offensive Technologies</i> (WOOT &rsquo;14), San Diego, CA, August 2014</div>
1405      <div class="abstract">
1406        <p>We introduce optimizations to the ZMap network scanner that achieve a
1407        10-fold increase in maximum scan rate. By parallelizing address
1408        generation, introducing an improved blacklisting algorithm, and using
1409        zero-copy NIC access, we drive ZMap to nearly the maximum throughput of
1410        10 gigabit Ethernet, almost 15 million probes per second. With these
1411        changes, ZMap can comprehensively scan for a single TCP port across the
1412        entire public IPv4 address space in 4.5 minutes given adequate upstream
1413        bandwidth. We consider the implications of such rapid scanning for both
1414        defenders and attackers, and we briefly discuss a range of potential
1415        applications.</p>
1416      </div>
1417    </div>
1418
1419    <div class="citation">
1420      <div class="title"><a href="https://fc14.ifca.ai/papers/fc14_submission_158.pdf">Elliptic Curve Cryptography in Practice</a> <span class="altlink">(<a href="https://eprint.iacr.org/2013/734">ePrint</a>)</span></div>
1421      <div class="author">Joppe&nbsp;W.&nbsp;Bos, J.&nbsp;Alex&nbsp;Halderman, Nadia&nbsp;Heninger, Jonathan&nbsp;Moore, Michael&nbsp;Naehrig, and Eric&nbsp;Wustrow</div>
1422      <div class="ref"><i>18th International Conference on Financial Cryptography and Data Security</i> (FC &rsquo;14), March 2014</div>
1423      <div class="abstract">
1424        <p>In this paper, we perform a review of elliptic curve cryptography
1425        (ECC), as it is used in practice today, in order to reveal unique
1426        mistakes and vulnerabilities that arise in implementations of ECC. We
1427        study four popular protocols that make use of this type of public-key
1428        cryptography: Bitcoin, secure shell (SSH), transport layer security
1429        (TLS), and the Austrian e-ID card. We are pleased to observe that about
1430        1 in 10 systems support ECC across the TLS and SSH protocols. However,
1431        we find that despite the high stakes of money, access and resources
1432        protected by ECC, implementations suffer from vulnerabilities similar to
1433        those that plague previous cryptographic systems.</p>
1434      </div>
1435    </div>
1436
1437    <div class="citation">
1438      <div class="title"><a href="/pub/papers/smartwatch13.pdf">Outsmarting Proctors with Smartwatches: A Case Study on Wearable Computing Security</a> <span class="altlink">(<a href="https://freedom-to-tinker.com/blog/amigi/cheating-on-exams-with-smartwatches/">blog post</a>)</span></div>
1439      <div class="author">Alex&nbsp;Migicovsky, Zakir&nbsp;Durumeric, Jeff&nbsp;Ringenberg, and J.&nbsp;Alex&nbsp;Halderman</div>
1440      <div class="ref"><i>18th International Conference on Financial Cryptography and Data Security</i> (FC &rsquo;14), March 2014</div>
1441      <div class="abstract">
1442        <p>Many companies have recently started to offer wearable computing
1443        devices including glasses, bracelets, and watches. While this technology
1444        enables exciting new applications, it also poses new security and
1445        privacy concerns. In this work, we explore these implications and
1446        analyze the impact of one of the first networked wearable
1447        devices&mdash;smartwatches&mdash;on an academic environment. As a proof
1448        of concept, we develop an application for the Pebble smartwatch called
1449        ConTest that would allow dishonest students to inconspicuously
1450        collaborate on multiple-choice exams in real time, using a cloud-based
1451        service, a smartphone, and a client application on the watch. We discuss
1452        the broader implications of this technology, suggest hardware and
1453        software approaches that can be used to prevent such attacks, and pose
1454        questions for future research.</p>
1455      </div>
1456    </div>
1457
1458    <div class="citation">
1459      <div class="title"><a href="/pub/papers/https-imc13.pdf">Analysis of the HTTPS Certificate Ecosystem</a></div>
1460      <div class="author">Zakir&nbsp;Durumeric, James&nbsp;Kasten, Michael&nbsp;Bailey, and J.&nbsp;Alex&nbsp;Halderman</div>
1461      <div class="ref"><i>13th ACM Internet Measurement Conference</i> (IMC &rsquo;13), Barcelona, Spain, October 2013</div>
1462      <div class="abstract">
1463        <p>We report the results of a large-scale measurement study of the HTTPS
1464        certificate ecosystem&mdash;the public-key infrastructure that underlies
1465        nearly all secure web communications. Using data collected by performing
1466        110 Internet-wide scans over 14 months, we gain detailed and temporally
1467        fine-grained visibility into this otherwise opaque area of
1468        security-critical infrastructure. We investigate the trust relationships
1469        among root authorities, intermediate authorities, and the leaf
1470        certificates used by web servers, ultimately identifying and classifying
1471        more than 1,800 entities that are able to issue certificates vouching
1472        for the identity of any website. We uncover practices that may put the
1473        security of the ecosystem at risk, and we identify frequent
1474        configuration problems that lead to user-facing errors and potential
1475        vulnerabilities. We conclude with lessons and recommendations to ensure
1476        the long-term health and security of the certificate ecosystem.</p>
1477      </div>
1478    </div>
1479
1480    <div class="citation">
1481      <div class="title"><a href="/pub/papers/zmap-sec13.pdf">ZMap: Fast Internet-Wide Scanning and its Security Applications</a> <span class="altlink">(<a href="https://www.zmap.io">website and tool</a>)</span></div>
1482      <div class="author">Zakir&nbsp;Durumeric, Eric&nbsp;Wustrow, and J.&nbsp;Alex&nbsp;Halderman</div>
1483      <div class="ref"><i>22nd USENIX Security Symposium</i> (Sec &rsquo;13), Washington, DC, August 2013</div>
1484      <div class="abstract">
1485        <p>Internet-wide network scanning has numerous security applications,
1486        including exposing new vulnerabilities and tracking the adoption of
1487        defensive mechanisms, but probing the entire public address space with
1488        existing tools is both difficult and slow. We introduce ZMap, a modular,
1489        open-source network scanner specifically architected to perform
1490        Internet-wide scans and capable of surveying the entire IPv4 address
1491        space in under 45 minutes from user space on a single machine,
1492        approaching the theoretical maximum speed of gigabit Ethernet. We
1493        present the scanner architecture, experimentally characterize its
1494        performance and accuracy, and explore the security implications of high
1495        speed Internet-scale network surveys, both offensive and defensive. We
1496        also discuss best practices for good Internet citizenship when
1497        performing Internet-wide surveys, informed by our own experiences
1498        conducting a long-term research survey over the past year.</p>
1499      </div>
1500    </div>
1501
1502    <div class="citation">
1503      <div class="title"><a href="https://jhalderm.com/pub/papers/ipmi-woot13.pdf">Illuminating the Security Issues Surrounding Lights-Out Server Management</a></div>
1504      <div class="author">Anthony&nbsp;Bonkoski, Russ&nbsp;Bielawski, and J.&nbsp;Alex&nbsp;Halderman</div>
1505      <div class="ref"><i>7th USENIX Workshop on Offensive Technologies</i> (WOOT &rsquo;13), Washington, DC, August 2013</div>
1506      <div class="abstract">
1507        <p>Out-of-band, lights-out management has become a standard feature on
1508        many servers, but while this technology can be a boon for system
1509        administrators, it also presents a new and interesting vector for
1510        attack. This paper examines the security implications of the Intelligent
1511        Platform Management Interface (IPMI), which is implemented on server
1512        motherboards using an embedded Baseboard Management Controller (BMC). We
1513        consider the threats posed by an incorrectly implemented IPMI and
1514        present evidence that IPMI vulnerabilities may be widespread. We analyze
1515        a major OEM&rsquo;s IPMI implementation and discover that it is riddled
1516        with textbook vulnerabilities, some of which would allow a remote
1517        attacker to gain root access to the BMC and potentially take control of
1518        the host system. Using data from Internet-wide scans, we find that there
1519        are at least 100,000 IPMI-enabled servers (across three large vendors)
1520        running on publicly accessible IP addresses, contrary to recommended
1521        best practice. Finally, we suggest defensive strategies for servers
1522        currently deployed and propose avenues for future work.</p>
1523      </div>
1524    </div>
1525
1526    <div class="citation">
1527      <div class="title"><a href="https://jhalderm.com/pub/papers/iran-foci13.pdf">Internet Censorship in Iran: A First Look</a></div>
1528      <div class="author">Simurgh&nbsp;Aryan, Homa&nbsp;Aryan, and J.&nbsp;Alex&nbsp;Halderman</div>
1529      <div class="ref"><i>3rd USENIX Workshop on Free and Open Communications on the Internet</i> (FOCI &rsquo;13), Washington, DC, August 2013</div>
1530      <div class="abstract">
1531        <p>The Iranian government operates one of the largest and most
1532        sophisticated Internet censorship regimes in the world, but the
1533        mechanisms it employs have received little research attention, primarily
1534        due to lack of access to network connections within the country and
1535        personal risks to Iranian citizens who take part. In this paper, we
1536        examine the status of Internet censorship in Iran based on network
1537        measurements conducted from a major Iranian ISP during the lead up to
1538        the June 2013 presidential election. We measure the scope of the
1539        censorship by probing Alexa&rsquo;s top 500 websites in 18 different
1540        categories. We investigate the technical mechanisms used for HTTP
1541        Host&ndash;based blocking, keyword filtering, DNS hijacking, and
1542        protocol-based throttling. Finally, we map the network topology of the
1543        censorship infrastructure and find evidence that it relies heavily on
1544        centralized equipment, a property that might be fruitfully exploited by
1545        next generation approaches to censorship circumvention.</p>
1546      </div>
1547    </div>
1548
1549    <div class="citation">
1550      <div class="title"><a href="https://jhalderm.com/pub/papers/cage-fc13.pdf">CAge: Taming Certificate Authorities by Inferring Restricted Scopes</a> <span class="altlink">(<a href="https://jhalderm.com/pub/papers/cage-fc13-ext.pdf">extended paper</a>)</span></div>
1551      <div class="author">James&nbsp;Kasten, Eric&nbsp;Wustrow, and J.&nbsp;Alex&nbsp;Halderman</div>
1552      <div class="ref"><i>17th International Conference on Financial Cryptography and Data Security</i> (FC &rsquo;13), Okinawa, Japan, April 2013</div>
1553      <div class="abstract">
1554        <p>The existing HTTPS public-key infrastructure (PKI) uses a
1555        coarse-grained trust model: either a certificate authority (CA) is
1556        trusted by browsers to vouch for the identity of <i>any</i> domain or it
1557        is not trusted at all. More than a thousand root and intermediate CAs
1558        can currently sign certificates for any domain and be trusted by popular
1559        browsers. This violates the principle of least privilege and creates an
1560        excessively large attack surface, as highlighted by recent CA
1561        compromises. In this paper, we present CAge, a mechanism that browser
1562        makers can apply to drastically reduce the excessive trust placed in CAs
1563        without fundamentally altering the CA ecosystem or breaking existing
1564        practice. CAge works by imposing restrictions on the set of top-level
1565        domains (TLDs) for which each CA is trusted to sign. Our key
1566        observation, based on an Internet-wide survey of TLS certificates, is
1567        that CAs commonly sign for only a handful of TLDs; in fact, 90% of CAs
1568        have signed certificates for domains in fewer than ten TLDs, and only
1569        35% have ever signed a certificate for a domain in <span
1570        class="tt">.com</span>. We show that it is possible to algorithmically
1571        <i>infer</i> reasonable restrictions on CAs&rsquo; trusted scopes based
1572        on this behavior, and we present evidence that browser-enforced inferred
1573        scopes would be a durable and effective way to reduce the attack surface
1574        of the HTTPS PKI. We find that simple inference rules can reduce the
1575        attack surface by nearly a factor of ten without hindering 99% of CA
1576        signing activity over a six-month period.</p>
1577      </div>
1578    </div>
1579
1580    <div class="citation">
1581      <div class="title"><a href="/pub/papers/weakkeys-sec12.pdf">Mining Your Ps and Qs: Detection of Widespread Weak Keys in Network Devices</a> <span class="altlink">(<a href="https://factorable.net">website</a>, <a href="/pub/papers/weakkeys-sec12-extended.pdf">extended paper</a>)</span></div>
1582      <div class="author">Nadia&nbsp;Heninger, Zakir&nbsp;Durumeric, Eric&nbsp;Wustrow, and J.&nbsp;Alex&nbsp;Halderman</div>
1583      <div class="ref"><i>21st USENIX Security Symposium</i> (Sec &rsquo;12), Bellevue, WA, August 2012 &mdash; <b>Best&nbsp;Paper&nbsp;Award</b> &mdash; <b>Test&nbsp;of&nbsp;Time&nbsp;Award (2022)</b></div>
1584      <div><small>Named one of Computing Reviews&rsquo; <a href="https://www.computingreviews.com/recommend/bestof/notableitems_2012.cfm">Notable Computing Books and Articles of 2012</a>.</small>&nbsp;</div>
1585      <div class="abstract">
1586        <p>RSA and DSA can fail catastrophically when used with malfunctioning
1587        random number generators, but the extent to which these problems arise
1588        in practice has never been comprehensively studied at Internet scale. We
1589        perform the largest ever network survey of TLS and SSH servers and
1590        present evidence that vulnerable keys are surprisingly widespread. We
1591        find that 0.75% of TLS certificates share keys due to insufficient
1592        entropy during key generation, and we suspect that another 1.70% come
1593        from the same faulty implementations and may be susceptible to
1594        compromise. Even more alarmingly, we are able to obtain RSA private keys
1595        for 0.50% of TLS hosts and 0.03% of SSH hosts, because their public keys
1596        shared nontrivial common factors due to entropy problems, and DSA
1597        private keys for 1.03% of SSH hosts, because of insufficient signature
1598        randomness. We cluster and investigate the vulnerable hosts, finding
1599        that the vast majority appear to be headless or embedded devices. In
1600        experiments with three software components commonly used by these
1601        devices, we are able to reproduce the vulnerabilities and identify
1602        specific software behaviors that induce them, including a boot-time
1603        entropy hole in the Linux random number generator. Finally, we suggest
1604        defenses and draw lessons for developers, users, and the security
1605        community.</p>
1606      </div>
1607    </div>
1608
1609    <div class="citation">
1610      <div class="title"><a href="/pub/papers/dcvoting-fc12.pdf">Attacking the Washington, D.C. Internet Voting System</a> <span class="altlink">(<a href="https://freedom-to-tinker.com/blog/jhalderm/hacking-dc-internet-voting-pilot">blog post</a>, <a href="https://josephhall.org/nqb2/index.php/dccouncildvm">testimony</a>)</span></div>
1611      <div class="author">Scott&nbsp;Wolchok, Eric&nbsp;Wustrow, Dawn&nbsp;Isabel, and J.&nbsp;Alex&nbsp;Halderman</div>
1612      <div class="ref"><i>16th Intl. Conference on Financial Cryptography and Data Security</i> (FC &rsquo;12), Bonaire, February 2012</div>
1613      <div class="abstract">
1614        <p>In 2010, Washington, D.C. developed an Internet voting pilot project
1615        that was intended to allow overseas absentee voters to cast their
1616        ballots using a website. Prior to deploying the system in the general
1617        election, the District held a unique public trial: a mock election
1618        during which anyone was invited to test the system or attempt to
1619        compromise its security. This paper describes our experience
1620        participating in this trial. Within 48 hours of the system going live,
1621        we had gained near-complete control of the election server. We
1622        successfully changed every vote and revealed almost every secret ballot.
1623        Election officials did not detect our intrusion for nearly two business
1624        days&mdash;and might have remained unaware for far longer had we not
1625        deliberately left a prominent clue. This case study&mdash;the first (to
1626        our knowledge) to analyze the security of a government Internet voting
1627        system from the perspective of an attacker in a realistic pre-election
1628        deployment&mdash;attempts to illuminate the practical challenges of
1629        securing online voting as practiced today by a growing number of
1630        jurisdictions.</p>
1631      </div>
1632    </div>
1633
1634    <div class="citation">
1635      <div class="title"><a href="/pub/papers/telex-sec11.pdf">
1635Telex: Anticensorship in the Network Infrastructure</a> <span class="altlink">(<a href="https://telex.cc">website and software</a>, <a href="https://www.youtube.com/watch?v=a_2IDvTLz7I">video</a>)</span></div>
1636      <div class="author">Eric&nbsp;Wustrow, Scott&nbsp;Wolchok, Ian&nbsp;Goldberg, and J.&nbsp;Alex&nbsp;Halderman</div>
1637      <div class="ref"><i>20th USENIX Security Symposium</i> (Sec &rsquo;11), San Francisco, CA, August 2011 &mdash; <b>Runner-up for <a href="https://petsymposium.org/award/winners.php">2012 PET Award</a></b></div>
1638      <div class="abstract">
1639        <p>In this paper, we present Telex, a new approach to resisting
1640        state-level Internet censorship. Rather than attempting to win the
1641        cat-and-mouse game of finding open proxies, we leverage censors&rsquo;
1642        unwillingness to completely block day-to-day Internet access. In effect,
1643        Telex converts innocuous, unblocked websites into proxies, without their
1644        explicit collaboration. We envision that friendly ISPs would deploy
1645        Telex stations on paths between censors&rsquo; networks and popular,
1646        uncensored Internet destinations. Telex stations would monitor seemingly
1647        innocuous flows for a special &ldquo;tag&rdquo; and transparently divert
1648        them to a forbidden website or service instead. We propose a new
1649        cryptographic scheme based on elliptic curves for tagging TLS handshakes
1650        such that the tag is visible to a Telex station but not to a censor. In
1651        addition, we use our tagging scheme to build a protocol that allows
1652        clients to connect to Telex stations while resisting both passive and
1653        active attacks. We also present a proof-of-concept implementation that
1654        demonstrates the feasibility of our system.</p>
1655      </div>
1656    </div>
1657
1658    <div class="citation">
1659      <div class="title"><a href="/pub/papers/china-pam11.pdf">Internet Censorship in China: Where Does the Filtering Occur?</a></div>
1660      <div class="author">Xueyang&nbsp;Xu, Z.&nbsp;Morley&nbsp;Mao, and J.&nbsp;Alex&nbsp;Halderman</div>
1661      <div class="ref"><i>12th Passive and Active Measurement Conference</i> (PAM &rsquo;11), Atlanta, GA, March 2011</div>
1662      <div class="abstract">
1663        <p>China filters Internet traffic in and out of the country. In order to
1664        circumvent the firewall, it is helpful to know where the filtering
1665        occurs. In this work, we explore the AS-level topology of China&rsquo;s
1666        network, and probe the firewall to find the locations of filtering
1667        devices. We find that even though most filtering occurs in border ASes,
1668        choke points also exist in many provincial networks. The result suggests
1669        that two major ISPs in China have different approaches for placing
1670        filtering devices.</p>
1671      </div>
1672    </div>
1673
1674    <div class="citation">
1675      <div class="title"><a href="/pub/papers/pwnage-fc11.pdf">Absolute Pwnage: Security Risks of Remote Administration Tools</a> <span class="altlink">(<a href="https://www.freedom-to-tinker.com/blog/jhalderm/schools-laptop-spying-software-exploitable-anywhere">blog post</a>)</span></div>
1676      <div class="author">Jay&nbsp;Novak, Jonathan&nbsp;Stribley, Kenneth&nbsp;Meagher, and J.&nbsp;Alex&nbsp;Halderman</div>
1677      <div class="ref"><i>15th International Financial Cryptography Conference</i> (FC &rsquo;11), February 2011</div>
1678      <div class="abstract">
1679        <p>Many IT departments use remote administration products to configure,
1680        monitor, and maintain the systems they manage. These tools can be
1681        beneficial in the right hands, but they can also be devastating if
1682        attackers exploit them to seize control of machines. As a case study, we
1683        analyze the security of a remote administration product called Absolute
1684        Manage. We find that the system&rsquo;s communication protocol suffers
1685        from serious design flaws and fails to provide adequate integrity,
1686        confidentiality, or authentication. Attackers can exploit these
1687        vulnerabilities to issue unauthorized commands on client systems and
1688        execute arbitrary code with administrator privileges. These blatant
1689        vulnerabilities suggest that remote administration tools require
1690        increased scrutiny from the security community. We recommend that
1691        developers adopt defensive designs that limit the damage attackers can
1692        cause if they gain control.</p>
1693      </div>
1694    </div>
1695
1696    <div class="citation">
1697      <div class="title"><a href="/pub/papers/voting-wecsr11.pdf">Ethical Issues in E-Voting Security Analysis</a></div>
1698      <div class="author">David&nbsp;G.&nbsp;Robinson and J.&nbsp;Alex&nbsp;Halderman</div>
1699      <div class="ref"><i>2nd Workshop on Ethics in Computer Security Research</i> (WECSR &rsquo;11), March 2011</div>
1700      <div class="abstract">
1701        <p>Research about weaknesses in deployed electronic voting systems
1702        raises a variety of pressing ethical concerns. In addition to ethical
1703        issues common to vulnerability research, such as the potential harms and
1704        benefits of vulnerability disclosure, electronic voting researchers face
1705        questions that flow from the unique and important role voting plays in
1706        modern democratic societies. Should researchers worry that their own
1707        work (not unlike the flaws they study) could sway an election outcome?
1708        When elected officials authorize a security review, how should
1709        researchers address the conflicted interests of these incumbent
1710        politicians, who may have powerful incentives to downplay problems, and
1711        might in principle be in a position to exploit knowledge about
1712        vulnerabilities when they stand for re-election? How should researchers
1713        address the risk that identifying specific flaws will lead to a false
1714        sense of security, after those particular problems have been resolved?
1715        This paper makes an early effort to address these and other questions
1716        with reference to experience from previous e-voting security reviews. We
1717        hope our provisional analysis will help practicing researchers
1718        anticipate and address ethical issues in future studies.</p>
1719      </div>
1720    </div>
1721
1722    <div class="citation">
1723      <div class="title"><a href="/pub/papers/evm-ccs10.pdf">Security Analysis of India&rsquo;s Electronic Voting Machines</a> <span class="altlink">(<a href="https://indiaevm.org">
1723website and video</a>)</span></div>
1724      <div class="author">Scott&nbsp;Wolchok, Eric&nbsp;Wustrow, J.&nbsp;Alex&nbsp;Halderman, Hari&nbsp;K.&nbsp;Prasad, Arun&nbsp;Kankipati, Sai&nbsp;Krishna&nbsp;Sakhamuri, Vasavya&nbsp;Yagati, and Rop&nbsp;Gonggrijp</div>
1725      <div class="ref"><i>17th ACM Conference on Computer and Communications Security</i> (CCS &rsquo;10), Chicago, IL, October 2010</div>
1726      <div class="abstract">
1727        <p>Elections in India are conducted almost exclusively using electronic
1728        voting machines developed over the past two decades by a pair of
1729        government-owned companies. These devices, known in India as EVMs, have
1730        been praised for their simple design, ease of use, and reliability, but
1731        recently they have also been criticized following widespread reports of
1732        election irregularities. Despite this criticism, many details of the
1733        machines&rsquo; design have never been publicly disclosed, and they have
1734        not been subjected to a rigorous, independent security evaluation. In
1735        this paper, we present a security analysis of a real Indian EVM obtained
1736        from an anonymous source. We describe the machine&rsquo;s design and
1737        operation in detail, and we evaluate its security in light of relevant
1738        election procedures. We conclude that in spite of the machines&rsquo;
1739        simplicity and minimal software trusted computing base, they are
1740        vulnerable to serious attacks that can alter election results and
1741        violate the secrecy of the ballot. We demonstrate two attacks,
1742        implemented using custom hardware, which could be carried out by
1743        dishonest election insiders or other criminals with only brief physical
1744        access to the machines. This case study carries important lessons for
1745        Indian elections and for electronic voting security more generally.</p>
1746      </div>
1747    </div>
1748
1749    <div class="citation">
1750      <div class="title"><a href="/pub/papers/dht-woot10.pdf">Crawling BitTorrent DHTs for Fun and Profit</a></div>
1751      <div class="author">Scott&nbsp;Wolchok and J.&nbsp;Alex&nbsp;Halderman</div>
1752      <div class="ref"><i>4th USENIX Workshop on Offensive Technologies</i> (WOOT &rsquo;10), Washington, DC, August 2010</div>
1753      <div class="abstract">
1754        <p>This paper presents two kinds of attacks based on crawling the DHTs
1755        used for distributed BitTorrent tracking. First, we show how pirates can
1756        use crawling to rebuild BitTorrent search engines just a few hours after
1757        they are shut down <i>(crawling for fun)</i>. Second, we show how
1758        content owners can use related techniques to monitor pirates&rsquo;
1759        behavior in preparation for legal attacks and negate any perceived
1760        anonymity of the decentralized BitTorrent architecture <i>(crawling for
1761        profit)</i>.</p>
1762        <p>We validate these attacks and measure their performance with a
1763        crawler we developed for the Vuze DHT. We find that we can establish a
1764        search engine with over one million torrents in under two hours using a
1765        single desktop PC. We also track 7.9 million IP addresses downloading
1766        1.5 million torrents over 16 days. These results imply that shifting
1767        from centralized BitTorrent tracking to DHT-based tracking will have
1768        mixed results for the file sharing arms race. While it will likely make
1769        illicit torrents harder to quash, it will not help users hide their
1770        activities.</p>
1771      </div>
1772    </div>
1773
1774    <div class="citation">
1775      <div class="title"><a href="/pub/papers/sketcha-www10.pdf">Sketcha: A Captcha Based on Line Drawings of 3D Models</a> <span class="altlink">(<a href="https://www.cs.princeton.edu/gfx/pubs/Ross_2010_SAC/index.php">website</a>)</span></div>
1776      <div class="author">Steven&nbsp;A.&nbsp;Ross, J.&nbsp;Alex&nbsp;Halderman, and Adam&nbsp;Finkelstein</div>
1777      <div class="ref"><i>19th International World Wide Web Conference</i> (WWW &rsquo;10), Raleigh, NC, April 2010</div>
1778      <div class="abstract">
1779        <p>This paper introduces a captcha based on upright orientation of line
1780        drawings rendered from 3D models. The models are selected from a large
1781        database, and images are rendered from random viewpoints, affording many
1782        different drawings from a single 3D model. The captcha presents the user
1783        with a set of images, and the user must choose an upright orientation
1784        for each image. This task generally requires understanding of the
1785        semantic content of the image, which is believed to be difficult for
1786        automatic algorithms. We describe a process called covert filtering
1787        whereby the image database can be continually refreshed with drawings
1788        that are known to have a high success rate for humans, by inserting
1789        randomly into the captcha new images to be evaluated. Our analysis shows
1790        that covert filtering can ensure that captchas are likely to be solvable
1791        by humans while deterring attackers who wish to learn a portion of the
1792        database. We performed several user studies that evaluate how
1793        effectively people can solve the captcha. Comparing these results to an
1794        attack based on machine learning, we find that humans possess a
1795        substantial performance advantage over computers.</p>
1796      </div>
1797    </div>
1798
1799    <div class="citation">
1800      <div class="title"><a href="/pub/papers/unvanish-ndss10-web.pdf">Defeating Vanish with Low-Cost Sybil Attacks Against Large DHTs</a> <span class="altlink">(<a href="https://www.freedom-to-tinker.com/blog/felten/breaking-vanish-story-security-research-action">blog post</a>)</span></div>
1801      <div class="author">Scott&nbsp;Wolchok, Owen&nbsp;S.&nbsp;Hofmann, Nadia&nbsp;Heninger, Edward&nbsp;W.&nbsp;Felten, J.&nbsp;Alex&nbsp;Halderman, Christopher&nbsp;J.&nbsp;Rossbach, Brent&nbsp;Waters, and Emmett&nbsp;Witchel</div>
1802      <div class="ref"><i>17th Network and Distributed System Security Symposium</i> (NDSS &rsquo;10), San Diego, CA, February-March 2010</div>
1803      <div class="abstract">
1804        <p>Researchers at the University of Washington recently proposed Vanish,
1805        a system for creating messages that automatically
1806        &ldquo;self-destruct&rdquo; after a period of time. Vanish works by
1807        encrypting each message with a random key and storing shares of the key
1808        in a large, public distributed hash table (DHT). Normally, DHTs expunge
1809        data older than a certain age. After they expire, the key is permanently
1810        lost, and the encrypted data is permanently unreadable. Vanish is an
1811        interesting approach to an important privacy problem, but, in its
1812        current form, it is insecure. In this paper, we defeat the deployed
1813        Vanish implementation, explain how the original paper&rsquo;s security
1814        analysis is flawed, and draw lessons for future system designs.</p>
1815        <p>We present two Sybil attacks against the current Vanish
1816        implementation, which stores its encryption keys in the million-node
1817        Vuze BitTorrent DHT. These attacks work by continuously crawling the DHT
1818        and saving each stored value before it ages out. They can efficie
1818ntly
1819        recover keys for more than 99% of Vanish messages. We show that the
1820        dominant cost of these attacks is network data transfer, not memory
1821        usage as the Vanish authors expected, and that the total cost is two
1822        orders of magnitude less than they estimated. While we consider
1823        potential defenses, we conclude that public DHTs like Vuze probably
1824        cannot provide strong security for Vanish.</p>
1825      </div>
1826    </div>
1827
1828    <div class="citation">
1829      <div class="title"><a href="/pub/papers/avc-evt09.pdf">Can DREs Provide Long-Lasting Security? The Case of Return-Oriented Programming and the AVC Advantage</a> <span class="altlink">(<a href="https://web.archive.org/web/20100625163538/https://cseweb.ucsd.edu/groups/security/avc/">website and video</a>)</span></div>
1830      <div class="author">Steve&nbsp;Checkoway, Ariel&nbsp;J.&nbsp;Feldman, Brian&nbsp;Kantor, J.&nbsp;Alex&nbsp;Halderman, Edward&nbsp;W.&nbsp;Felten, and Hovav&nbsp;Shacham</div>
1831      <div class="ref"><i>2009 USENIX/ACCURATE/IAVoSS Electronic Voting Technology Workshop</i> (EVT &rsquo;09), Montreal, QC, August 2009</div>
1832      <div class="abstract">
1833        <p>A secure voting machine design must withstand new attacks devised
1834        throughout its multidecade service lifetime. In this paper, we give a
1835        case study of the longterm security of a voting machine, the Sequoia AVC
1836        Advantage, whose design dates back to the early 80s. The AVC Advantage
1837        was designed with promising security features: its software is stored
1838        entirely in read-only memory and the hardware refuses to execute
1839        instructions fetched from RAM. Nevertheless, we demonstrate that an
1840        attacker can induce the AVC Advantage to misbehave in arbitrary
1841        ways&mdash;including changing the outcome of an election&mdash;by means
1842        of a memory cartridge containing a specially-formatted payload. Our
1843        attack makes essential use of a recently-invented exploitation technique
1844        called return-oriented programming, adapted here to the Z80 processor.
1845        In return-oriented programming, short snippets of benign code already
1846        present in the system are combined to yield malicious behavior. Our
1847        results demonstrate the relevance of recent ideas from systems security
1848        to voting machine research, and vice versa. We had no access either to
1849        source code or documentation beyond that available on Sequoia&rsquo;s
1850        web site. We have created a complete vote-stealing demonstration exploit
1851        and verified that it works correctly on the actual hardware.</p>
1852      </div>
1853    </div>
1854
1855    <div class="citation">
1856      <div class="title"><a href="/pub/papers/paper-oak09.pdf">Fingerprinting Blank Paper Using Commodity Scanners</a> <span class="altlink">(<a href="https://web.archive.org/web/20110116061554/http://citp.princeton.edu/paper/">website</a>)</span></div>
1857      <div class="author">William&nbsp;Clarkson, Tim&nbsp;Weyrich, Adam&nbsp;Finkelstein, Nadia&nbsp;Heninger, J.&nbsp;Alex&nbsp;Halderman, and Edward&nbsp;W.&nbsp;Felten</div>
1858      <div class="ref"><i>30th IEEE Symposium on Security and Privacy</i> (Oakland &rsquo;09), Oakland, CA, May 2009</div>
1859      <div class="abstract">
1860        <p>This paper presents a novel technique for authenticating physical
1861        documents based on random, naturally occurring imperfections in paper
1862        texture. We introduce a new method for measuring the three-dimensional
1863        surface of a page using only a commodity scanner and without modifying
1864        the document in any way. From this physical feature, we generate a
1865        concise fingerprint that uniquely identifies the document. Our technique
1866        is secure against counterfeiting and robust to harsh handling; it can be
1867        used even before any content is printed on a page. It has a wide range
1868        of applications, including detecting forged currency and tickets,
1869        authenticating passports, and halting counterfeit goods. Document
1870        identification could also be applied maliciously to de-anonymize printed
1871        surveys and to compromise the secrecy of paper ballots.</p>
1872      </div>
1873    </div>
1874
1875    <div class="citation">
1876      <div class="title"><a href="/pub/papers/coldboot-sec08.pdf">Lest We Remember: Cold Boot Attacks on Encryption Keys</a> <span class="altlink">(<a href="https://web.archive.org/web/20110429202434/http://citp.princeton.edu/memory/">website and videos</a>)</span></div>
1877      <div class="author">J.&nbsp;Alex&nbsp;Halderman, Seth&nbsp;D.&nbsp;Schoen, Nadia&nbsp;Heninger, William&nbsp;Clarkson, William&nbsp;Paul, Joseph&nbsp;A.&nbsp;Calandrino, Ariel&nbsp;J.&nbsp;Feldman, Jacob&nbsp;Appelbaum, and Edward&nbsp;W.&nbsp;Felten</div>
1878      <div class="ref"><i>17th USENIX Security Symposium</i> (Sec &rsquo;08), San Jose, CA, July 2008 &mdash; <b>Best&nbsp;Student&nbsp;Paper&nbsp;Award</b> &mdash; <b><a href="https://web.archive.org/web/20210506040522/https://pwnie
1878s.com/archive/2008/winners/">Pwnie&nbsp;Award&nbsp;</a></b></div>
1879      <div class="ref">Reprinted in <i><a href="/pub/papers/coldboot-cacm09.pdf">Communications of the ACM</a></i>, <!--52(5):91-98,--> May 2009</div>
1880      <div class="abstract">
1881        <p>Contrary to popular assumption, DRAMs used in most modern computers
1882        retain their contents for seconds to minutes after power is lost, even
1883        at operating temperatures and even if removed from a motherboard.
1884        Although DRAMs become less reliable when they are not refreshed, they
1885        are not immediately erased, and their contents persist sufficiently for
1886        malicious (or forensic) acquisition of usable full-system memory images.
1887        We show that this phenomenon limits the ability of an operating system
1888        to protect cryptographic key material from an attacker with physical
1889        access. We use cold reboots to mount attacks on popular disk encryption
1890        systems &#8212; BitLocker, FileVault, dm-crypt, and TrueCrypt &#8212;
1891        using no special devices or materials. We experimentally characterize
1892        the extent and predictability of memory remanence and report that
1893        remanence times can be increased dramatically with simple techniques. We
1894        offer new algorithms for finding cryptographic keys in memory images and
1895        for correcting errors caused by bit decay. Though we discuss several
1896        strategies for partially mitigating these risks, we know of no simple
1897        remedy that would eliminate them.</p>
1898      </div>
1899    </div>
1900
1901    <div class="citation">
1902      <div class="title"><a href="/pub/papers/prng-evt08.pdf">In Defense of Pseudorandom Sample Selection</a></div>
1903      <div class="author">Joseph&nbsp;A.&nbsp;Calandrino, J.&nbsp;Alex&nbsp;Halderman, and Edward&nbsp;W.&nbsp;Felten</div>
1904      <div class="ref"><i>2008 USENIX/ACCURATE Electronic Voting Technology Workshop</i> (EVT &rsquo;08), San Jose, CA, July 2008</div>
1905      <div class="abstract">
1906        <p>Generation of random numbers is a critical component of existing
1907        post-election auditing techniques. Recent work has largely discouraged
1908        the use of all pseudorandom number generators, including
1909        cryptographically secure pseudorandom number generators (CSPRNGs), for
1910        this purpose, instead recommending the sole use of observable physical
1911        techniques. In particular, simple dice rolling has received a great deal
1912        of positive attention. The typical justification for this recommendation
1913        is that those less comfortable with mathematics prefer a simple,
1914        observable technique. This paper takes a contrary view. Simple,
1915        observable techniques like dice rolling are not necessarily robust
1916        against sleight of hand and other forms of fraud, and attempts to harden
1917        them against fraud can dramatically increase their complexity. With
1918        simple dice rolling, we know of no techniques that provide citizens with
1919        a reasonable means of verifying that fraud did not occur during the roll
1920        process. CSPRNGs, used properly, can be simple, robust, and verifiable,
1921        and they allow for the use of auditing techniques that might otherwise
1922        be impractical. While we understand initial skepticism towards this
1923        option, we argue that appropriate use of CSPRNGs would strengthen audit
1924        security.</p>
1925      </div>
1926    </div>
1927
1928    <div class="citation">
1929      <div class="title"><a href="/pub/papers/stopgap-evt08.pdf">You Go to Elections with the Voting System You Have: Stop-Gap Mitigations for Deployed Voting Systems</a></div>
1930      <div class="author">J.&nbsp;Alex&nbsp;Halderman, Eric&nbsp;Rescorla, Hovav&nbsp;Shacham, and David&nbsp;Wagner</div>
1931      <div class="ref"><i>2008 USENIX/ACCURATE Electronic Voting Technology Workshop</i> (EVT &rsquo;08), San Jose, CA, July 2008</div>
1932      <div class="abstract">
1933        <p>In light of the systemic vulnerabilities uncovered by recent reviews
1934        of deployed e-voting systems, the surest way to secure the voting
1935        process would be to scrap the existing systems and design new ones.
1936        Unfortunately, engineering new systems will take years, and many
1937        jurisdictions are unlikely to be able to afford new equipment in the
1938        near future. In this paper we ask how jurisdictions can make the best
1939        use of the equipment they already own until they can replace it.
1940        Starting from current practice, we propose defenses that involve new but
1941        realistic procedures, modest changes to existing software, and no
1942        changes to existing hardware. Our techniques achieve greatly improved
1943        protection against outsider attacks: they provide containment of viral
1944        spread, improve the integrity of vote tabulation, and offer some
1945        detection of individual compromised devices. They do not provide
1946        security against insiders with access to election management systems,
1947        which appears to require significantly greater changes to the existing
1948        systems.</p>
1949      </div>
1950    </div>
1951
1952    <div class="citation">
1953      <div class="title"><a href="/pub/papers/harvest-ccs07.pdf">Harvesting Verifiable Challenges from Oblivious Online Sources</a></div>
1954      <div class="author">J.&nbsp;Alex&nbsp;Halderman and Brent&nbsp;Waters</div>
1955      <div class="ref"><i>14th ACM Conference on Computer and Communications Security</i> (CCS &rsquo;07), Washington, DC, October 2007</div>
1956      <div class="abstract">
1957        <p>Several important security protocols require parties to perform
1958        computations based on random challenges. Traditionally, proving that the
1959        challenges were randomly chosen has required interactive communication
1960        among the parties or the existence of a trusted server. We offer an
1961        alternative solution where challenges are harvested from oblivious
1962        servers on the Internet. This paper describes a framework for deriving
1963        &ldquo;harvested challenges&rdquo; by mixing data from various
1964        pre-existing online sources. While individual sources may become
1965        predictable or fall under adversarial control, we provide a policy
1966        language that allows application developers to specify comb
1966inations of
1967        sources that meet their security needs. Participants can then convince
1968        each other that their challenges were formed freshly and in accordance
1969        with the policy. We present Combine, an open source implementation of
1970        our framework, and show how it can be applied to a variety of
1971        applications, including remote storage auditing and non-interactive
1972        client puzzles.</p>
1973      </div>
1974    </div>
1975
1976    <div class="citation">
1977      <div class="title"><a href="/pub/papers/audit-evt07-full.pdf">Machine-Assisted Election Auditing</a> <span class="altlink">(<a href="/pub/papers/audit-evt07.pdf">workshop version</a>)</span></div>
1978      <div class="author">Joseph&nbsp;A.&nbsp;Calandrino, J.&nbsp;Alex&nbsp;Halderman, and Edward&nbsp;W.&nbsp;Felten</div>
1979      <div class="ref"><i>2007 USENIX/ACCURATE Electronic Voting Technology Workshop</i> (EVT &rsquo;07), Boston, MA, August 2007</div>
1980      <div class="abstract">
1981        <p>Election audit procedures usually rely on precinct based recounts, in
1982        which workers manually review all paper ballots from selected polling
1983        places, but these recounts can be expensive due to the labor required.
1984        This paper proposes an alternative audit strategy that allows machines
1985        to perform most of the work. Precincts are recounted using recounting
1986        machines, and their output is manually audited using efficient ballot
1987        sampling techniques. This strategy can achieve equal or greater
1988        confidence than precinct-based auditing at a significantly lower cost
1989        while protecting voter privacy better than previous ballot-based
1990        auditing methods. We show how to determine which ballots to audit
1991        against the recounting machines&rsquo; records and compare this new
1992        approach to precinct-based audits in the context of Virginia&rsquo;s
1993        November 2006 election. Far fewer ballots need to be audited by hand
1994        using our approach. We also explore extensions to these techniques, such
1995        as varying individual ballots&rsquo; audit probabilities based on the
1996        votes they contain, that promise further efficiency gains.</p>
1997      </div>
1998    </div>
1999
2000    <div class="citation">
2001      <div class="title"><a href="/pub/papers/ts-evt07.pdf">Security Analysis of the Diebold AccuVote-TS Voting Machine</a> <span class="altlink">(<a href="/pub/papers/ts-evt07-init.pdf">initial public version</a>, <a href="https://web.archive.org/web/20110713050334/http://citp.princeton.edu/voting/">website and videos</a>)</span></div>
2002      <div class="author">Ariel&nbsp;J.&nbsp;Feldman, J.&nbsp;Alex&nbsp;Halderman, and Edward&nbsp;W.&nbsp;Felten</div>
2003      <div class="ref"><i>2007 USENIX/ACCURATE Electronic Voting Technology Workshop</i> (EVT &rsquo;07), Boston, MA, August 2007</div>
2004      <div class="abstract">
2005        <p>This paper presents a fully independent security study of a Diebold
2006        AccuVote-TS voting machine, including its hardware and software. We
2007        obtained the machine from a private party. Analysis of the machine, in
2008        light of real election procedures, shows that it is vulnerable to
2009        extremely serious attacks. For example, an attacker who gets physical
2010        access to a machine or its removable memory card for as little as one
2011        minute could install malicious code; malicious code on a machine could
2012        steal votes undetectably, modifying all records, logs, and counters to
2013        be consistent with the fraudulent vote count it creates. An attacker
2014        could also create malicious code that spreads automatically and silently
2015        from machine to machine during normal election activities &mdash; a
2016        voting-machine virus. We have constructed working demonstrations of
2017        these attacks in our lab. Mitigating these threats will require changes
2018        to the voting machine&rsquo;s hardware and software and the adoption of
2019        more rigorous election procedures.</p>
2020      </div>
2021    </div>
2022
2023    <div class="citation">
2024      <div class="title"><a href="/pub/papers/rootkit-sec06.pdf">Lessons from the Sony CD DRM Episode</a> <span class="altlink">(<a href="/pub/papers/rootkit-sec06-full.pdf">extended version</a>, <a href="https://web.archive.org/web/20120128072803/https://freedom-to-tinker.com/tags/cd-copy-protection">blog posts</a>)</span></div>
2025      <div class="author">J.&nbsp;Alex&nbsp;Halderman and Edward&nbsp;W.&nbsp;Felten</div>
2026      <div class="ref"><i>15th USENIX Security Symposium</i> (Sec &rsquo;06), Vancouver, BC, August 2006</div>
2027      <div class="abstract">
2028        <p>In the fall of 2005, problems discovered in two Sony-BMG compact disc
2029        copy protection systems, XCP and MediaMax, triggered a public uproar
2030        that ultimately led to class-action litigation and the recall of
2031        millions of discs. We present an in-depth analysis of these
2032        technologies, including their design, implementation, and deployment.
2033        The systems are surprisingly complex and suffer from a diverse array of
2034        flaws that weaken their content protection and expose users to serious
2035        security and privacy risks. Their complexity, and their failure, makes
2036        them an interesting case study of digital rights management that carries
2037        valuable lessons for content companies, DRM vendors, policymakers, end
2038        users, and the security community.</p>
2039      </div>
2040    </div>
2041
2042    <div class="citation">
2043      <div class="title"><a href="/pub/papers/password-www05.pdf">A Convenient Method for Securely Managing Passwords</a></div>
2044      <div class="author">J.&nbsp;Alex&nbsp;Halderman, Brent&nbsp;Waters, and Edward&nbsp;W.&nbsp;Felten</div>
2045      <div class="ref"><i>14th International World Wide Web Conference</i> (WWW &rsquo;05), Chiba, Japan, May 2005</div>
2046      <div class="abstract">
2047        <p>Computer users are asked to generate, keep secret, and recall an
2048        increasing number of passwords for uses including host accounts, email
2049        servers, e-commerce sites, and online financial services. Unfortunately,
2050        the password entropy that users can comfortably memorize seems
2051        insufficient to store unique, secure passwords for all these accounts,
2052        and it is likely to remain constant as the number of passwords (and the
2053        adversary&rsquo;s computational power) increases into the future. In
2054        this paper, we propose a technique that uses a strengthened
2055        cryptographic hash function to compute secure passwords for arbitrarily
2056        many accounts while requiring the user to memorize only a single short
2057        password. This mechanism functions entirely on the client; no
2058        server-side changes are needed. Unlike previous approaches, our design
2059        is both highly resistant to brute force attacks and nearly stateless,
2060        allowing users to retrieve their passwords from any location so long as
2061        they can execute our program and remember a short secret. This
2062        combination of security and convenience will, we believe, entice users
2063        to adopt our scheme. We discuss the construction of our algorithm in
2064        detail, compare its strengths and weaknesses to those of related
2065        approaches, and present Password Multiplier, an implementation in the
2066        form of an extension to the Mozilla Firefox web browser.</p>
2067      </div>
2068    </div>
2069
2070    <div class="citation">
2071      <div class="title"><a href="/pub/papers/memex-wpes04.pdf">Privacy Management for Portable Recording Devices</a></div>
2072      <div class="author">J.&nbsp;Alex&nbsp;Halderman, Brent&nbsp;Waters, and Edward&nbsp;W.&nbsp;Felten</div>
2073      <div class="ref"><i>2004 ACM Workshop on Privacy in the Electronic Society</i> (WPES &rsquo;04), Washington, DC, October 2004</div>
2074      <div class="abstract">
2075        <p>The growing popularity of inexpensive, portable recording devices,
2076        such as cellular phone cameras and compact digital audio recorders,
2077        presents a significant new threat to privacy. We propose a set of
2078        technologies that can be integrated into recording devices to provide
2079        stronger, more accurately targeted privacy protections than other legal
2080        and technical measures now under consideration. Our design is based on
2081        an informed consent principle, which it supports by the use of novel
2082        devices and protocols that automate negotiations over consent and ensure
2083        appropriate safeguards on recorded data. We define the protocols needed
2084        for this purpose and establish their security. We also describe a
2085        working prototype implementation that safeguards audio recorded by
2086        laptop PCs in a wireless network.</p>
2087      </div>
2088    </div>
2089
2090    <div class="citation">
2091      <div class="title"><a href="/pub/papers/puzzle-ccs04.pdf">New Client Puzzle Outsourcing Techniques for DoS Protection</a></div>
2092      <div class="author">Brent&nbsp;Waters, Ari&nbsp;Juels, J.&nbsp;Alex&nbsp;Halderman, and Edward&nbsp;W.&nbsp;Felten</div>
2093      <div class="ref"><i>11th ACM Conference on Computer and Communications Security</i> (CCS &rsquo;04), Washington, DC, October 2004</div>
2094      <div class="abstract">
2095        <p>We explore new techniques for the use of cryptographic puzzles as a
2096        countermeasure to Denial-of-Service (DoS) attacks. We propose simple new
2097        techniques that permit the <i>outsourcing</i> of puzzles--their
2098        distribution via a robust external service that we call a
2099        <i>bastion</i>. Many servers can rely on puzzles distributed by a single
2100        bastion. We show how a bastion, somewhat surprisingly, need not know
2101        which servers rely on its services. Indeed, in one of our constructions,
2102        a bastion may consist merely of a publicly accessible random data
2103        source, rather than a special purpose server. Our outsourcing techniques
2104        help eliminate puzzle distribution as a point of compromise. Our design
2105        has three main advantages over prior approaches. First, it is more
2106        resistant to DoS attacks aimed at the puzzle mechanism itself,
2107        withstanding over 80% more attack traffic than previous methods in our
2108        experiments. Second, our scheme is cheap enough to apply at the IP
2109        level, though it also works at higher levels of the protocol stack.
2110        Third, our method allows clients to solve puzzles offline, reducing the
2111        need for users to wait while their computers solve puzzles. We present a
2112        prototype implementation of our approach, and we describe experiments
2113        that validate our performance claims.</p>
2114      </div>
2115    </div>
2116
2117    <div class="citation">
2118      <div class="title"><a href="/pub/papers/search-web3d03.pdf">Early Experiences with a 3D Model Search Engine</a></div>
2119      <div class="author">Patrick&nbsp;Min, J.&nbsp;Alex&nbsp;Halderman, Michael&nbsp;Kazhdan, and Thomas&nbsp;A.&nbsp;Funkhouser</div>
2120      <div class="ref"><i>8th International Conference on 3D Web Technology</i> (Web3D &rsquo;03), March 2003 &mdash; <b>Best&nbsp;Paper&nbsp;Award</b></div>
2121      <div class="abstract">
2122        <p>New acquisition and modeling tools make it easier to create 3D
2123        models, and fordable and powerful graphics hardware makes it easier to
2124        use them. As a result, the number of 3D models available on the web is
2125        increasing rapidly. However, it is still not as easy to find 3D models
2126        as it is to find, for example, text documents and images. What is needed
2127        is a &ldquo;3D model search engine,&rdquo; a specialized search engine
2128        that targets 3D models. We created a prototype 3D model search engine to
2129        investigate the design and implementation issues. Our search engine can
2130        be partitioned into three main components: (1) acquisition: 3D models
2131        have to be collected from the web, (2) analysis: they have to be
2132        analyzed for later matching, and (3) query processing and matching: an
2133        online system has to match user queries to the collected 3D models. Our
2134        site currently indexes over 36,000 models, of which about 31,000 are
2135        freely available. In addition to a text search interface, it offers
2136        several 3D and 2D shape-based query interfaces. Since it went online one
2137        year ago (in November 2001), it has processed over 148,000 searches from
2138        37,800 hosts in 103 different countries. Currently 20-25% of the about
2139        1,000 visitors per week are returning users. This paper reports on our
2140        initial experiences designing, building, and running the 3D model search
2141        engine.</p>
2142      </div>
2143    </div>
2144
2145    <div class="citation">
2146      <div class="title"><a href="/pub/papers/search-tog03.pdf">A Search Engine for 3D Models</a></div>
2147      <div class="author">Thomas&nbsp;Funkhouser, Patrick&nbsp;Min, Misha&nbsp;Kazhdan, Joyce&nbsp;Chen, J.&nbsp;Alex&nbsp;Halderman, David&nbsp;Dobkin, and David&nbsp;Jacobs</div>
2148      <div class="ref"><i>ACM Transactions on Graphics</i>, 22(1):83-105, January 2003</div>
2149      <div class="abstract">
2150        <p>As the number of 3D models available on the Web grows, there is an
2151        increasing need for a search engine to help people find them.
2152        Unfortunately, traditional text-based search techniques are not always
2153        effective for 3D data. In this paper, we investigate new shape-based
2154        search methods. The key challenges are to develop query methods simple
2155        enough for novice users and matching algorithms robust enough to work
2156        for arbitrary polygonal models. We present a web-based search engine
2157        system that supports queries based on 3D sketches, 2D sketches, 3D
2158        models, and/or text keywords. For the shape-based queries, we have
2159        developed a new matching algorithm that uses spherical harmonics to
2160        compute discriminating similarity measures without requiring repair of
2161        model degeneracies or alignment of orientations. It provides 46.245%
2162        better performance than related shape matching methods during
2163        precision-recall experiments, and it is fast enough to return query
2164        results from a repository of 20,000 models in under a second. The net
2165        result is a growing interactive index of 3D models available on the Web
2166        (i.e., a Google for 3D models).</p>
2167      </div>
2168    </div>
2169
2170    <div class="citation">
2171      <div class="title"><a href="/pub/papers/cdcopy-drm02.pdf">Evaluating New Copy-Prevention Techniques for Audio CDs</a></div>
2172      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2173      <div class="ref"><i>ACM Workshop on Digital Rights Management</i> (DRM &rsquo;02), Washington, DC, November 2002</div>
2174      <div class="abstract">
2175        <p>Several major record labels are adopting a new family of
2176        copy-prevention techniques intended to limit &ldquo;casual&rdquo;
2177        copying by compact disc owners using their personal computers. These
2178        employ deliberate data errors introduced into discs during manufacturing
2179        to cause incompatibility with PCs without affecting ordinary CD players.
2180        We examine three such recordings: A Tribute to Jim Reeves by Charley
2181        Pride, A New Day Has Come by Celine Dion, and More Music from The Fast
2182        and the Furious by various artists. In tests with different CD-ROM
2183        drives, operating systems, and playback software, we find these discs
2184        are unreadable in several widely-used applications as of July 2002. We
2185        analyze the specific technical differences between the modified
2186        recordings and standard audio CDs, and we consider repairs to hardware
2187        and software that would restore compatibility. We conclude that these
2188        schemes are harmful to legitimate CD owners and will not reduce illegal
2189        copying in the long term, so the music industry should reconsider their
2190        deployment.</p>
2191      </div>
2192    </div>
2193
2194    <h3>Selected other publications</h3>
2195
2196    <div class="citation">
2197      <div class="title"><a href="/pub/misc/eac-comment-2023.pdf">Voter Privacy and VVSG 2.0</a></div>
2198      <div class="author">Braden&nbsp;Crimmins, Dhanya&nbsp;Narayanan, J.&nbsp;Alex&nbsp;Halderman, and Drew&nbsp;Springall</div>
2199      <div class="ref">Public comments in response to <a href="https://www.regulations.gov/document/EAC-2023-0001-0001">U.S. EAC VVSG 2.0 annual review</a>, June 6, 2023</div>
2200      <div class="spacer"><!-- No abstract --></div>
2201    </div>
2202
2203    <div class="citation">
2204      <div class="title"><a href="/pub/misc/mi-hb4210-testimony-2023.pdf">Testimony in Opposition to Internet Voting in Michigan</a></div>
2205      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2206      <div class="ref">Written testimony to the <a href="https://www.legislature.mi.gov/Committees/Meeting?meetingID=fc7e896f46444806ab6aefaf6b573128">Michigan House Committee on Elections</a> regarding <a href="https://www.legislature.mi.gov/Bills/Bill?ObjectName=2023-HB-4210">H.B. 4210</a>, May 9, 2023</div>
2207      <div class="spacer"><!-- No abstract --></div>
2208    </div>
2209
2210    <div class="citation">
2211      <div class="title">
2211<a href="/pub/misc/pde-memorial-2023.pdf">Remembering Peter Eckersley</a></div>
2212      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2213      <div class="ref">Remarks delivered at his <a href="https://ai.objectives.institute/pde-memorial">memorial service</a>, March 4, 2023</div>
2214      <div class="spacer"><!-- No abstract --></div>
2215    </div>
2216
2217    <div class="citation">
2218      <div class="title"><a href="https://dvsorder.org">The DVSorder Vulnerability</a></div>
2219      <div class="author">Braden&nbsp;Crimmins, Dhanya&nbsp;Narayanan, Josiah&nbsp;Walker, Drew&nbsp;Springall, and J.&nbsp;Alex&nbsp;Halderman</div>
2220      <div class="ref">Oct. 2022</div>
2221      <div class="spacer"><!-- No abstract --></div>
2222    </div>
2223
2224    <div class="citation">
2225      <div class="title"><a href="https://www.cisa.gov/uscert/ics/advisories/icsa-22-154-01">ICS Advisory: Vulnerabilities Affecting Dominion Voting Systems ImageCast X</a></div>
2226      <div class="author">Findings credited to J.&nbsp;Alex&nbsp;Halderman and Drew&nbsp;Springall</div>
2227      <div class="ref">CISA/ICS-CERT (ICSA-22-154-01), June 3, 2022</div>
2228      <div class="abstract">
2229        <p>This advisory identifies vulnerabilities affecting versions of the
2230        Dominion Voting Systems Democracy Suite ImageCast X, which is an
2231        in-person voting system used to allow voters to mark their ballot. The
2232        ImageCast X can be configured to allow a voter to produce a paper record
2233        or to record votes electronically. While these vulnerabilities present
2234        risks that should be mitigated as soon as possible, CISA has no evidence
2235        that these vulnerabilities have been exploited in any elections.</p>
2236        <p>Exploitation of these vulnerabilities would require physical access
2237        to individual ImageCast X devices, access to the Election Management
2238        System (EMS), or the ability to modify files before they are uploaded to
2239        ImageCast X devices. Jurisdictions can prevent and/or detect the
2240        exploitation of these vulnerabilities by diligently applying the
2241        mitigations recommended in this advisory, including technical, physical,
2242        and operational controls that limit unauthorized access or manipulation
2243        of voting systems. Many of these mitigations are already typically
2244        standard practice in jurisdictions where these devices are in use and
2245        can be enhanced to further guard against exploitation of these
2246        vulnerabilities.</p>
2247      </div>
2248    </div>
2249
2250    <div class="citation">
2251      <div class="title"><a href="https://www.newsweek.com/election-security-problems-still-must-addressed-opinion-1632592">Election Security Problems Still Must Be Addressed</a></div>
2252      <div class="author">Susan&nbsp;Greenhalgh and J.&nbsp;Alex&nbsp;Halderman</div>
2253      <div class="ref"><i>Newsweek</i>, September 27, 2021</div>
2254      <div class="spacer"><!-- No abstract --></div>
2255    </div>
2256
2257    <div class="citation">
2258      <div class="title"><a href="https://storage.courtlistener.com/recap/gov.uscourts.gand.240678/gov.uscourts.gand.240678.1681.0.pdf">Security Analysis of Georgia&rsquo;s ImageCast X Ballot Marking Devices</a> <span class="altlink">(<a class="external" href="https://freedom-to-tinker.com/2023/06/14/security-analysis-of-the-dominion-imagecast-x/">blog post</a>; <a class="external" href="https://www.cisa.gov/uscert/ics/advisories/icsa-22-154-01">CISA advisory</a>)</span></div>
2259      <div class="author">J.&nbsp;Alex&nbsp;Halderman and Drew&nbsp;Springall</div>
2260      <div class="ref">Expert report submitted on behalf of plaintiffs Donna Curling, et al. in <i>Curling v. Raffensperger</i>, Civil Action No. 1:17-CV-2989-AT, U.S. District Court for the Northern District of Georgia, Atlanta Division, July 1, 2021</div>
2261      <div class="spacer"><!-- No abstract --></div>
2262    </div>
2263
2264    <div class="citation">
2265      <div class="title"><a href="https://www.michigan.gov/documents/sos/Antrim_720623_7.pdf">Analysis of the Antrim County, Michigan November 2020 Election Incident</a></div>
2266      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2267      <div class="ref">Expert report prepared for the State of Michigan, March 26, 2021</div>
2268      <div class="spacer"><!-- No abstract --></div>
2269    </div>
2270
2271    <div class="citation">
2272      <div class="title"><a href="https://www.barrons.com/articles/elections-should-be-grounded-in-evidence-not-blind-trust-51609769710">Elections Should be Grounded in Evidence, Not Blind Trust</a></div>
2273      <div class="author">Philip&nbsp;B.&nbsp;Stark, Edward&nbsp;Perez, and J.&nbsp;Alex&nbsp;Halderman</div>
2274      <div class="ref"><i>Barrons</i>, January 4, 2021</div>
2275      <div class="spacer"><!-- No abstract --></div>
2276    </div>
2277
2278    <div class="citation">
2279      <div class="title"><a href="https://www.michigan.gov/documents/sos/ESAC_Report_Recommendations_706522_7.pdf">Michigan Election Security Advisory Commission Report and Recommendations</a></div>
2280      <div class="author">J.&nbsp;Alex&nbsp;Halderman <i>et al.</i></div>
2281      <div class="ref">Report prepared for the State of Michigan, October 2020</div>
2282      <div class="spacer"><!-- No abstract --></div>
2283    </div>
2284
2285    <div class="citation">
2286      <div class="title"><a href="https://slate.com/technology/2020/01/internet-voting-could-destroy-our-elections.html">Internet Voting Is Happening Now&mdash;And it could destroy our elections</a></div>
2287      <div class="author">Rachel&nbsp;Goodman and J.&nbsp;Alex&nbsp;Halderman</div>
2288      <div class="ref"><i>Slate</i>, January 15, 2020</div>
2289      <div class="spacer"><!-- No abstract --></div>
2290    </div>
2291
2292    <div class="citation">
2293      <div class="title"><a href="/pub/misc/fsgg-voting-written19.pdf">Congressional Testimony Regarding Federal Funding for Election Cybersecurity</a><span class="altlink">(<a href="/pub/misc/fsgg-voting-written19.pdf">written statement</a>, <a href="/pub/misc/fsgg-voting-oral19.pdf">oral statement</a>, <a href="https://www.youtube.com/watch?v=-6nybiIC-9k">video</a>)</span></div>
2294      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2295      <div class="ref">Testimony before the U.S. House Appropriations Subcommittee on Financial Service and General Government, <a href="https://web.archive.org/web/20210118003255/https://appropriations.house.gov/legislation/hearings/election-security-ensuring-the-integrity-of-us-election-systems">Election Security: Ensuring the Integrity of U.S. Election Systems</a>, February 27, 2019</div>
2296      <div class="spacer"><!-- No abstract --></div>
2297    </div>
2298
2299    <div class="citation">
2300      <div class="title">
2300<a href="https://www.nytimes.com/2018/04/05/opinion/election-voting-machine-hacking-russians.html">I Hacked an Election. So Can the Russians.</a></div>
2301      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2302      <div class="ref">Video op/ed in collaboration with <i>The New York Times</i>, April 5, 2018</div>
2303      <div class="spacer"><!-- No abstract --></div>
2304    </div>
2305
2306    <div class="citation">
2307      <div class="title"><a href="/pub/misc/ssci-voting-testimony17.pdf">Congressional Testimony Regarding Russian Interference in the 2016 U.S. Elections</a> <span class="altlink">(<a href="https://www.youtube.com/watch?v=AmivIHUAy8Q#t=2h15m13s">video</a></span>)</div>
2308      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2309      <div class="ref">Testimony before the U.S. Senate Select Committee on Intelligence, June 21, 2017</div>
2310      <div class="spacer"><!-- No abstract --></div>
2311    </div>
2312
2313    <div class="citation">
2314      <div class="title"><a href="https://www.washingtonpost.com/news/posteverything/wp/2017/06/21/heres-how-to-keep-russian-hackers-from-attacking-the-2018-elections">Here&rsquo;s How to Keep Russian Hackers from Attacking the 2018 Elections</a></div>
2315      <div class="author">J.&nbsp;Alex&nbsp;Halderman and Justin&nbsp;Talbot-Zorn</div>
2316      <div class="ref"><i>The Washington Post</i>, June 21, 2017</div>
2317      <div class="spacer"><!-- No abstract --></div>
2318    </div>
2319
2320    <div class="citation">
2321      <div class="title"><a href="/pub/papers/ch7-evoting-attacks-2016.pdf">Practical Attacks on Real-world E-voting</a></div>
2322      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2323      <div class="ref">In Feng Hao and Peter Y. A. Ryan (Eds.), <i>Real-World Electronic Voting: Design, Analysis and Deployment</i>, pages 145&ndash;171, CRC Press, December 2016</div>
2324      <div class="spacer"><!-- No abstract --></div>
2325    </div>
2326
2327    <div class="citation">
2328      <div class="title"><a href="https://medium.com/@jhalderm/want-to-know-if-the-election-was-hacked-look-at-the-ballots-c61a6113b0ba">Want to Know if the Election was Hacked? Look at the Ballots</a></div>
2329      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2330      <div class="ref">Posted on Medium, November 23, 2016. (Read by over a million people.)</div>
2331      <div class="spacer"><!-- No abstract --></div>
2332    </div>
2333
2334    <div class="citation">
2335      <div class="title"><a href="https://spectrum.ieee.org/tech-talk/telecom/security/the-security-challenges-of-online-voting-have-not-gone-away">The Security Challenges of Online Voting Have Not Gone Away</a></div>
2336      <div class="author">Robert&nbsp;Cunningham, Matthew&nbsp;Bernhard, and J.&nbsp;Alex&nbsp;Halderman</div>
2337      <div class="ref"><i>IEEE Spectrum</i>, November 3, 2016</div>
2338      <div class="spacer"><!-- No abstract --></div>
2339    </div>
2340
2341    <div class="citation">
2342      <div class="title"><a href="https://www.eecs.umich.edu/techreports/cse/2014/CSE-TR-586-14.pdf">TIVOS: Trusted Visual I/O Paths for Android</a></div>
2343      <div class="author">Earlence&nbsp;Fernandes, Qi&nbsp;Alfred&nbsp;Chen, Georg&nbsp;Essl, J.&nbsp;Alex&nbsp;Halderman, Z.&nbsp;Morley&nbsp;Mao, and Atul&nbsp;Prakash</div>
2344      <div class="ref"><i>University of Michigan Technical Report</i>, May 2014</div>
2345      <div class="abstract">
2346        <p>Stealthy pixel-perfect attacks on smartphone apps are a class of
2347        phishing attacks that rely on visual deception to trick users into
2348        entering sensitive information into trojan apps. We introduce an
2349        operating system abstraction called Trusted Visual I/O Paths (TIVOs)
2350        that enables a user to securely verify the app she is interacting with,
2351        only assuming that the operating system provides a trusted computing
2352        base. As proof of concept, we built a TIVO for Android, one that is
2353        activated any time a soft keyboard is used by an application (e.g., for
2354        password entry) so that the user can reliably determine the app that
2355        receives the user&rsquo;s keyboard input. We implemented TIVO by
2356        modifying Android&rsquo;s user-interface stack and evaluated the
2357        abstraction using a controlled user study where users had to decide
2358        whether to trust the login screen of four different applications that
2359        were randomly subjected to two forms of pixel-perfect attacks. The TIVO
2360        mechanism was found to significantly reduce the effectiveness of
2361        pixel-perfect attacks, with acceptable impact on overall usability and
2362        only modest performance overhead.</p>
2363      </div>
2364    </div>
2365
2366    <div class="citation">
2367      <div class="title"><a href="https://www.foreignaffairs.com/articles/140214/nadia-heninger-and-j-alex-halderman/tales-from-the-crypto-community">Tales from the Crypto Community: The NSA Hurt Cybersecurity. Now It Should Come Clean</a></div>
2368      <div class="author">Nadia&nbsp;Heninger and J.&nbsp;Alex&nbsp;Halderman</div>
2369      <div class="ref"><i>Foreign Affairs</i>, October 23, 2013</div>
2370      <div class="abstract">
2371        <p>Of all of the revelations about the NSA that have come to light in
2372        recent months, two stand out as the most worrisome and surprising to
2373        cybersecurity experts. The first is that the NSA has worked to weaken
2374        the international cryptographic standards that define how computers
2375        secure communications and data. The second is that the NSA has
2376        deliberately introduced backdoors into security-critical software and
2377        hardware. If the NSA has indeed engaged in such activities, it has
2378        risked the computer security of the United States (and the world) as
2379        much as any malicious attacks have to date.</p>
2380      </div>
2381    </div>
2382
2383    <div class="citation">
2384      <div class="title"><a href="https://ieeexplore.ieee.org/document/5439535">To Strengthen Security, Change Developers&rsquo; Incentives</a></div>
2385      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2386      <div class="ref"><i>IEEE Security and Privacy</i>, March/April 2010</div>
2387      <div class="abstract">
2388        <p>Many common software vulnerabilities are avoidable if software makers
2389        apply appropriate care, yet developers&rsquo; incentives often lead them
2390        to underinvest in security. Profit-maximizing developers invest to the
2391        extent that strengthening security increases sales or reduces their
2392        liability, yet these incentives are undermined by the softw
2392are
2393        market&rsquo;s structure. By understanding and reshaping such
2394        incentives, we can greatly improve security at comparably low cost. The
2395        author argues for requiring increased transparency about security
2396        problems and development practices, which will help software buyers make
2397        better-informed purchases, and for holding developers liable for the
2398        costs of security failures caused by their products.</p>
2399      </div>
2400    </div>
2401
2402    <div class="citation">
2403      <div class="title"><a href="/pub/gd/">Analysis of the Green Dam Censorware System</a></div>
2404      <div class="author">Scott&nbsp;Wolchok, Randy&nbsp;Yao, and J.&nbsp;Alex&nbsp;Halderman</div>
2405      <div class="ref"><i>University of Michigan Technical Report</i>, June 11, 2009</div>
2406      <div class="abstract">
2407        <p>We have discovered remotely-exploitable vulnerabilities in Green Dam,
2408        the censorship software reportedly mandated by the Chinese government.
2409        Any web site a Green Dam user visits can take control of the PC.
2410        According to press reports, China will soon require all PCs sold in the
2411        country to include Green Dam. This software monitors web sites visited
2412        and other activity on the computer and blocks adult content as well as
2413        politically sensitive material. We examined the Green Dam software and
2414        found that it contains serious security vulnerabilities due to
2415        programming errors. Once Green Dam is installed, any web site the user
2416        visits can exploit these problems to take control of the computer. This
2417        could allow malicious sites to steal private data, send spam, or enlist
2418        the computer in a botnet. In addition, we found vulnerabilities in the
2419        way Green Dam processes blacklist updates that could allow the software
2420        makers or others to install malicious code during the update process. We
2421        found these problems with less than 12 hours of testing, and we believe
2422        they may be only the tip of the iceberg. Green Dam makes frequent use of
2423        unsafe and outdated programming practices that likely introduce numerous
2424        other vulnerabilities. Correcting these problems will require extensive
2425        changes to the software and careful retesting. In the meantime, we
2426        recommend that users protect themselves by uninstalling Green Dam
2427        immediately.</p>
2428      </div>
2429    </div>
2430
2431    <div class="citation">
2432      <div class="title"><a href="/pub/papers/avc-tr08.pdf">AVC Advantage: Hardware Functional Specifications</a></div>
2433      <div class="author">J.&nbsp;Alex&nbsp;Halderman and Ariel&nbsp;J.&nbsp;Feldman</div>
2434      <div class="ref"><i>Princeton University Computer Science Technical Report TR-816-08</i>, March 8, 2008</div>
2435      <div class="abstract">
2436        <p>This report describes the hardware design of the AVC Advantage
2437        direct-recording electronic (DRE) voting machine. We developed these
2438        functional specifications by reverse engineering a government-surplus
2439        system.</p>
2440      </div>
2441    </div>
2442
2443    <div class="citation">
2444      <div class="title"><a href="/pub/papers/diebold-ttbr07.pdf">Source Code Review of the Diebold Voting System</a></div>
2445      <div class="author">Joseph&nbsp;A.&nbsp;Calandrino, Ariel&nbsp;J.&nbsp;Feldman, J.&nbsp;Alex&nbsp;Halderman, David&nbsp;Wagner, Harlan&nbsp;Yu, and William&nbsp;Zeller</div>
2446      <div class="ref">Part of the California Secretary of State&rsquo;s <a href="https://www.sos.ca.gov/elections/ovsta/frequently-requested-information/top-bottom-review">&ldquo;Top-to-Bottom&rdquo; Voting Systems Review</a>, July 2007</div>
2447      <div class="spacer"><!-- No abstract --></div>
2448    </div>
2449
2450    <div class="citation">
2451      <div class="title"><a href="/pub/papers/drm-sp06.pdf">Digital Rights Management, Spyware, and Security</a></div>
2452      <div class="author">Edward&nbsp;W.&nbsp;Felten and J.&nbsp;Alex&nbsp;Halderman</div>
2453      <div class="ref"><i>IEEE Security and Privacy</i>, January/February 2006</div>
2454      <div class="spacer"><!-- No abstract --></div>
2455    </div>
2456
2457    <div class="citation">
2458      <div class="title"><a href="/pub/cd3/">Analysis of the MediaMax CD3 Copy-Prevention System</a></div>
2459      <div class="author">J.&nbsp;Alex&nbsp;Halderman</div>
2460      <div class="ref"><i>Princeton University Computer Science Technical Report TR-679-03</i>, October 2003</div>
2461      <div class="abstract">
2462        <p>
2462MediaMax CD3 is a new copy-prevention technique from SunnComm
2463        Technologies that is designed to prevent unauthorized copying of audio
2464        CDs using personal computers. SunnComm claims its product facilitates
2465        &ldquo;a verifiable and commendable level of security,&rdquo; but in
2466        tests on a newly-released album, I find that the protections may have no
2467        effect on a large fraction of deployed PCs, and that most users who
2468        would be affected can bypass the system entirely by holding the shift
2469        key every time they insert the CD. I explain that MediaMax interferes
2470        with audio copying by installing a device driver the first time software
2471        from the CD is executed, but I show that this provides only minimal
2472        protection because the driver can easily be disabled. I also examine the
2473        digital rights management system used to control access to a set of
2474        encrypted, compressed audio files distributed on the CD. Although
2475        restrictions on these files are more relaxed than in prior copy
2476        protected discs, they still prohibit many uses permitted by the law. I
2477        conclude that MediaMax and similar copy-prevention systems are
2478        irreparably flawed but predict that record companies will find success
2479        with more customer-friendly alternatives for reducing infringement.</p>
2480      </div>
2481    </div>
2482  </div>
2483
2484</body>
2485</html>

Line numbers count LF bytes from the start of the resource, as the search results do. Vendor segments are library code the classifier recognised; they are stored but not indexed. Bytes are shown as Latin1 characters, one per byte.