Publications

As Internet censors rapidly evolve new blocking techniques, circumvention tools must also adapt and roll out new strategies to remain unblocked. But new strategies can be time consuming for circumventors to develop and deploy, and usually an update to one tool often requires significant additional effort to be ported to others. Moreover, distributing the updated application across different platforms poses its own set of challenges.

In this paper, we introduce WATER (WebAssembly Transport Executables Runtime), a novel design that enables applications to use a WebAssembly-based application-layer to wrap network transports (e.g., TLS). Deploying a new circumvention technique with WATER only requires distributing the WebAssembly Transport Module(WATM) binary and any transport-specific configuration, allowing dynamic transport updates without any change to the application itself. WATMs are also designed to be generic such that different applications using WATER can use the same WATM to rapidly deploy successful circumvention techniques to their own users, facilitating rapid interoperability between independent circumvention tools.

We explore the commonalities between methods for assuring the security of computer systems (cybersecurity) and the mechanisms that have evolved through natural selection to protect vertebrates against pathogens, and how insights derived from studying the evolution of natural defenses can inform the design of more effective cybersecurity systems. More generally, security challenges are crucial for the maintenance of a wide range of complex adaptive systems, including financial systems, and again lessons learned from the study of the evolution of natural defenses can provide guidance for the protection of such systems.

Pre-election logic and accuracy (L&A) testing is a process in which election officials validate the behavior of voting equipment by casting a known set of test ballots and confirming the expected results. Ideally, such testing can serve to detect certain forms of human error or fraud and help bolster voter confidence. We present the first detailed analysis of L&A testing practices across the United States. We find that while all states require L&A testing before every election, their implementations vary dramatically in scope, transparency, and rigorousness. We summarize each state’s requirements and score them according to uniform criteria. We also highlight best practices and flag opportunities for improvement, in hopes of encouraging broader adoption of more effective L&A processes.

In November 2020, Antrim County, Michigan published unofficial election results that misstated totals in the presidential race and other contests by up to several thousand votes. Antrim subsequently issued a series of corrections, and the certified presidential results were confirmed by a hand count. Nevertheless, Antrim was cited by the President of the United States as evidence of widespread fraud, and it remains a centerpiece of conspiracy theories about the 2020 election. At the request of the Michigan Secretary of State and Attorney General, I performed a forensic investigation of the incident. Using data from the election system, I precisely reproduce the major anomalies, explain their cause, and verify that they have been corrected. I also uncover other errors affecting specific down-ballot contests that have not been corrected, despite the unusual attention focused on the results, one of which may have changed the outcome of a local contest. Based on this analysis, I refute misinformation about the incident, concluding that it was not the result of a security breach but rather a series of operator errors compounded by inadequate procedures and insufficiently defensive software design. These events offer lessons for improving election administration and highlight the value of rigorously investigating election technology incidents for enhancing accuracy and public trust.

VPN adoption has seen steady growth over the past decade due to increased public awareness of privacy and surveillance threats. In response, certain governments are attempting to restrict VPN access by identifying connections using “dual use” DPI technology. To investigate the potential for VPN blocking, we develop mechanisms for accurately fingerprinting connections using OpenVPN, the most popular protocol for commercial VPN services. We identify three fingerprints based on protocol features such as byte pattern, packet size, and server response. Playing the role of an attacker who controls the network, we design a two-phase framework that performs passive fingerprinting and active probing in sequence. We evaluate our framework in partnership with a millionuser ISP and find that we identify over 85% of OpenVPN flows with only negligible false positives, suggesting that OpenVPN-based services can be effectively blocked with little collateral damage. Although some commercial VPNs implement countermeasures to avoid detection, our framework successfully identified connections to 34 out of 41 “obfuscated” VPN configurations. We discuss the implications of the VPN fingerprintability for different threat models and propose short-term defenses. In the longer term, we urge commercial VPN providers to be more transparent about their obfuscation approaches and to adopt more principled detection countermeasures, such as those developed in censorship circumvention research.

Postal voting is growing rapidly in the U.S., with 43% of voters casting ballots by mail in 2020, yet until recently there has been little research about extending the protections of end-to-end verifiable (E2E-V) election schemes to vote-by-mail contexts. The first—and to date, only—framework to focus on this setting is STROBE, which has important usability limitations. In this work, we present two approaches, RemoteVote and SAFE Vote, that allow mail-in voters to benefit from E2E-V without changing the voter experience for those who choose not to participate in verification. To evaluate these systems and compare them with STROBE, we consider an expansive set of properties, including novel attributes of usability and verifiability, several of which have applicability beyond vote-by-mail contexts. We hope that our work will help catalyze further progress towards universal applicability of E2E-V for real-world elections.

Most U.S. voters cast hand-marked paper ballots that are counted by optical scanners. Deployed ballot scanners typically utilize simplistic mark-detection methods, based on comparing the measured intensity of target areas to preset thresholds, but this technique is known to sometimes misread “marginal” marks that deviate from ballot instructions. We investigate the feasibility of improving scanner accuracy using supervised learning. We train a convolutional neural network to classify various styles of marks extracted from a large corpus of voted ballots. This approach achieves higher accuracy than a naive intensity threshold while requiring far fewer ballots to undergo manual adjudication. It is robust to imperfect feature extraction, as may be experienced in ballots that lack timing marks, and efficient enough to be performed in real time using contemporary central-count scanner hardware.

Democracy Live’s OmniBallot platform is a web-based system for blank ballot delivery, ballot marking, and online voting. In early 2020, three states—Delaware, West Virginia, and New Jersey—announced that they would allow certain voters to cast votes online using OmniBallot, but, despite the well established risks of Internet voting, the system has never before undergone a public, independent security review.

We reverse engineered the client-side portion of OmniBallot, as used in Delaware, in order to detail the system’s operation and analyze its security. We find that OmniBallot uses a simplistic approach to Internet voting that is vulnerable to vote manipulation by malware on the voter’s device and by insiders or other attackers who can compromise Democracy Live, Amazon, Google, or Cloudflare. In addition, Democracy Live, which had no privacy policy prior to our work, receives sensitive personally identifiable information—including the voter’s identity, ballot selections, and browser fingerprint—that could be used to target political ads or disinformation campaigns. Even when OmniBallot is used to mark ballots that will be printed and returned in the mail, the software sends the voter’s identity and ballot choices to Democracy Live, an unnecessary risk that jeopardizes the secret ballot.

We recommend changes to make the platform safer for ballot delivery and marking. However, we conclude that using OmniBallot for electronic ballot return represents a severe risk to election security and could allow attackers to alter election results without detection. In response to our findings, Delaware and New Jersey halted their use of OmniBallot for online voting, but it remains available in other jurisdictions, as do similar tools that likely face the same serious risks.

Increased adoption of HTTPS has created a largely encrypted web, but these security gains are on a collision course with governments that desire visibility into and control over user communications. Last year, the government of Kazakhstan conducted an unprecedented large-scale HTTPS interception attack by forcing users to trust a custom root certificate. We were able to detect the interception and monitor its scale and evolution using measurements from in-country vantage points and remote measurement techniques. We find that the attack targeted connections to 37 unique domains, with a focus on social media and communication services, suggesting a surveillance motive, and that it affected a large fraction of connections passing through the country’s largest ISP, Kazakhtelecom. Our continuous real-time measurements indicated that the interception system was shut down after being intermittently active for 21 days. Subsequently, supported by our findings, two major browsers (Mozilla Firefox and Google Chrome) completely blocked the use of Kazakhstan’s custom root. However, the incident sets a dangerous precedent, not only for Kazakhstan but for other countries that may seek to circumvent encryption online.

Refraction networking is a next-generation censorship circumvention approach that locates proxy functionality in the network itself, at participating ISPs or other network operators. Following years of research and development and a brief pilot, we established the world’s first production deployment of a Refraction Networking system. Our deployment uses a high-performance implementation of the TapDance protocol and is enabled as a transport in the popular circumvention app Psiphon. It uses TapDance stations at four physical uplink locations of a mid-sized ISP, Merit Network, with an aggregate bandwidth of 140 Gbps. By the end of 2019, our system was enabled as a transport option in 559,000 installations of Psiphon, and it served upwards of 33,000 unique users per month. This paper reports on our experience building the deployment and operating it for the first year. We describe how we overcame engineering challenges, present detailed performance metrics, and analyze how our system has responded to dynamic censor behavior. Finally, we review lessons learned from operating this unique artifact and discuss prospects for further scaling Refraction Networking to meet the needs of censored users.

Transnational Internet performance is an important indication of a country’s level of infrastructure investment, globalization, and openness. We conduct a large-scale measurement study of transnational Internet performance in and out of 29 countries and regions, and find six countries that have surprisingly low performance. Five of them are African countries and the last is mainland China, a significant outlier with major discrepancies between downstream and upstream performance. We then conduct a comprehensive investigation of the unusual transnational Internet performance of mainland China, which we refer to as the “Great Bottleneck of China”. Our results show that this bottleneck is widespread, affecting 79% of the receiver–sender pairs we measured. More than 70% of the pairs suffer from extremely slow speed (less than 1 Mbps) for more than 5 hours every day. In most tests the bottleneck appeared to be located deep inside China, suggesting poor network infrastructure to handle transnational traffic. The phenomenon has far-reaching implications for Chinese users’ browsing habits as well as for the ability of foreign Internet services to reach Chinese customers.

Ballot marking devices (BMDs) allow voters to select candidates on a computer kiosk, which prints a paper ballot that the voter can review before inserting it into a scanner to be tabulated. Unlike paperless voting machines, BMDs provide voters an opportunity to verify an auditable physical record of their choices, and a growing number of U.S. jurisdictions are adopting them for all voters. However, the security of BMDs depends on how reliably voters notice and correct any adversarially induced errors on their printed ballots. In order to measure voters’ error detection abilities, we conducted a large study (N=241) in a realistic polling place setting using real voting machines that we modified to introduce an error into each printout. Without intervention, only 40% of participants reviewed their printed ballots at all, and only 6.6% told a poll worker something was wrong. We also find that carefully designed interventions can improve verification performance. Verbally instructing voters to review the printouts and providing a written slate of candidates for whom to vote both significantly increased review and reporting rates—although the improvements may not be large enough to provide strong security in close elections, especially when BMDs are used by all voters. Based on these findings, we make several evidence-based recommendations to help better defend BMD-based elections.

Let’s Encrypt is a free, open, and automated HTTPS certificate authority (CA) created to advance HTTPS adoption to the entire Web. Since its launch in late 2015, Let’s Encrypt has grown to become the world’s largest HTTPS CA, accounting for more currently valid certificates than all other browser-trusted CAs combined. By January 2019, it had issued over 538 million certificates for 223 million domain names. We describe how we built Let’s Encrypt, including the architecture of the CA software system (Boulder) and the structure of the organization that operates it (ISRG), and we discuss lessons learned from the experience. We also describe the design of ACME, the IETF-standard protocol we created to automate CA-server interactions and certificate issuance, and survey the diverse ecosystem of ACME clients, including Certbot, a software agent we created to automate HTTPS deployment. Finally, we measure Let’s Encrypt’s impact on the Web and the CA ecosystem. We hope that the success of Let’s Encrypt can provide a model for further enhancements to the Web PKI and for future Internet security infrastructure.

Refraction Networking (formerly known as “Decoy Routing”) has emerged as a promising next-generation approach for circumventing Internet censorship. Rather than trying to hide individual circumvention proxy servers from censors, proxy functionality is implemented in the core of the network, at cooperating ISPs in friendly countries. Any connection that traverses these ISPs could be a conduit for the free flow of information, so censors cannot easily block access without also blocking many legitimate sites. While one Refraction scheme, TapDance, has recently been deployed at ISP-scale, it suffers from several problems: a limited number of “decoy” sites in realistic deployments, high technical complexity, and undesirable tradeoffs between performance and observability by the censor. These challenges may impede broader deployment and ultimately allow censors to block such techniques.

We present Conjure, an improved Refraction Networking approach that overcomes these limitations by leveraging unused address space at deploying ISPs. Instead of using real websites as the decoy destinations for proxy connections, our scheme connects to IP addresses where no web server exists leveraging proxy functionality from the core of the network. These phantom hosts are difficult for a censor to distinguish from real ones, but can be used by clients as proxies. We define the Conjure protocol, analyze its security, and evaluate a prototype using an ISP testbed. Our results suggest that Conjure can be harder to block than TapDance, is simpler to maintain and deploy, and offers substantially better network performance.

As paper ballots and post-election audits gain increased adoption in the United States, election technology vendors are offering products that allow jurisdictions to review ballot images—digital scans produced by optical-scan voting machines—in their post-election audit procedures. Jurisdictions including the state of Maryland rely on such image audits as an alternative to inspecting the physical paper ballots. We show that image audits can be reliably defeated by an attacker who can run malicious code on the voting machines or election management system. Using computer vision techniques, we develop an algorithm that automatically and seamlessly manipulates ballot images, moving voters’ marks so that they appear to be votes for the attacker’s preferred candidate. Our implementation is compatible with many widely used ballot styles, and we show that it is effective using a large corpus of ballot images from a real election. We also show that the attack can be delivered in the form of a malicious Windows scanner driver, which we test with a scanner that has been certified for use in vote tabulation by the U.S. Election Assistance Commission. These results demonstrate that post-election audits must inspect physical ballots, not merely ballot images, if they are to strongly defend against computer-based attacks on widely used voting systems.

HTTPS and TLS are the backbone of Internet security, however setting up web servers to run these protocols is a notoriously difficult process. In this paper, we perform two live subjects usability studies on the deployment of HTTPS in a real-world setting. Study 1 is a within subjects comparison between traditional HTTPS configuration (purchasing a certificate and installing it on a server) and Let’s Encrypt, which automates much of the process. Study 2 is a between subjects study looking at the same two systems, examining why users encounter usability issues. Overall we confirm past results that HTTPS is difficult to deploy, and we find some evidence that suggests Let’s Encrypt is an easier, more efficient method for deploying HTTPS.

We present a method and software for ballot-polling risk-limiting audits (RLAs) based on Bernoulli sampling: ballots are included in the sample with probability p, independently. Bernoulli sampling has several advantages: (1) it does not require a ballot manifest; (2) it can be conducted independently at different locations, rather than requiring a central authority to select the sample from the whole population of cast ballots or requiring stratified sampling; (3) it can start in polling places on election night, before margins are known. If the reported margins for the 2016 U.S. Presidential election are correct, a Bernoulli ballot-polling audit with a risk limit of 5% and a sampling rate of p₀=1% would have had at least a 99% probability of confirming the outcome in 42 states. (The other states were more likely to have needed to examine additional ballots.) Logistical and security advantages that auditing in the polling place affords may outweigh the cost of examining more ballots than some other methods might require.

We report the first wide-scale measurement study of serverside geographic restriction, or geoblocking, a phenomenon in which server operators intentionally deny access to users from particular countries or regions. Many sites practice geoblocking due to legal requirements or other business reasons, but excessive blocking can needlessly deny valuable content and services to entire national populations.

To help researchers and policymakers understand this phenomenon, we develop a semi-automated system to detect instances where whole websites were rendered inaccessible due to geoblocking. By focusing on detecting geoblocking capabilities offered by large CDNs and cloud providers, we can reliably distinguish the practice from dynamic anti-abuse mechanisms and network-based censorship. We apply our techniques to test for geoblocking across the Alexa Top 10K sites from thousands of vantage points in 177 countries. We then expand our measurement to a sample of CDN customers in the Alexa Top 1M.

We find that geoblocking occurs across a broad set of countries and sites. We observe geoblocking in nearly all countries we study, with Iran, Syria, Sudan, Cuba, and Russia experiencing the highest rates. These countries experience particularly high rates of geoblocking for finance and banking sites, likely as a result of U.S. economic sanctions. We also verify our measurements with data provided by Cloudflare, and find our observations to be accurate.

Remote censorship measurement tools can now detect DNS- and IP-based blocking at global scale. However, a major unmonitored form of interference is blocking triggered by deep packet inspection of application-layer data. We close this gap by introducing Quack, a scalable, remote measurement system that can efficiently detect application-layer interference.

We show that Quack can effectively detect application-layer blocking triggered on HTTP and TLS headers, and it is flexible enough to support many other diverse protocols. In experiments, we test for blocking across 4458 autonomous systems, an order of magnitude larger than provided by country probes used by OONI. We also test a corpus of 100,000 keywords from vantage points in 40 countries to produce detailed national blocklists. Finally, we analyze the keywords we find blocked to provide insight into the application-layer blocking ecosystem and compare countries’ behavior. We find that the most consistently blocked services are related to circumvention tools, pornography, and gambling, but that there is significant country-to-country variation.

Certificate Authorities (CAs) regularly make mechanical errors when issuing certificates. To quantify these errors, we introduce ZLint, a certificate linter that codifies the policies set forth by the CA/Browser Forum Baseline Requirements and RFC 5280 that can be tested in isolation. We run ZLint on browser-trusted certificates in Censys and systematically analyze how well CAs construct certificates. We find that the number errors has drastically reduced since 2012. In 2017, only 0.02% of certificates have errors. However, this is largely due to a handful of large authorities that consistently issue correct certificates. There remains a long tail of small authorities that regularly issue non-conformant certificates. We further find that issuing certificates with errors is correlated with other types of mismanagement and for large authorities, browser action. Drawing on our analysis, we conclude with a discussion on how the community can best use lint data to identify authorities with worrisome organizational practices and ensure long-term health of the Web PKI.

Internet access in Cuba is severely constrained, due to limited availability, slow speeds, and high cost. Within this isolated environment, technology enthusiasts have constructed a disconnected but vibrant IP network that has grown organically to reach tens of thousands of households across Havana. We present the first detailed characterization of this deployment, which is known as the SNET, or Street Network. Working in collaboration with SNET operators, we describe the network’s infrastructure and map its topology, and we measure bandwidth, available services, usage patterns, and user demographics. Qualitatively, we attempt to answer why the SNET exists and what benefits it has afforded its users. We go on to discuss technical challenges the network faces, including scalability, security, and organizational issues. To our knowledge, the SNET is the largest isolated community-driven network in existence, and its structure, successes, and obstacles show fascinating contrasts and similarities to those of the Internet at large.

Elections seem simple—aren’t they just counting? But they have a unique, challenging combination of security and privacy requirements. The stakes are high; the context is adversarial; the electorate needs to be convinced that the results are correct; and the secrecy of the ballot must be ensured. And they have practical constraints: time is of the essence, and voting systems need to be affordable and maintainable, and usable by voters, election officials, and pollworkers. It is thus not surprising that voting is a rich research area spanning theory, applied cryptography, practical systems analysis, usable security, and statistics. Election integrity involves two key concepts: convincing evidence that outcomes are correct and privacy, which amounts to convincing assurance that there is no evidence about how any given person voted. These are obviously in tension. We examine how current systems walk this tightrope.

The Mirai botnet, composed primarily of embedded and IoT devices, took the Internet by storm in late 2016 when it overwhelmed several high-profile targets with massive distributed denial-of-service (DDoS) attacks. In this paper, we provide a seven-month retrospective analysis of Mirai’s growth to a peak of 600k infections and a history of its DDoS victims. By combining a variety of measurement perspectives, we analyze how the botnet emerged, what classes of devices were affected, and how Mirai variants evolved and competed for vulnerable hosts. Our measurements serve as a lens into the fragile ecosystem of IoT devices. We argue that Mirai may represent a sea change in the evolutionary development of botnets—the simplicity through which devices were infected and its precipitous growth, demonstrate that novice malicious techniques can compromise enough low-end devices to threaten even some of the best-defended targets. To address this risk, we recommend technical and nontechnical interventions, as well as propose future research directions.

We report initial results from the world’s first ISP-scale field trial of a refraction networking system. Refraction networking is a next-generation censorship circumvention approach that locates proxy functionality in the middle of the network, at participating ISPs or other network operators. We built a high-performance implementation of the TapDance refraction networking scheme and deployed it on four ISP uplinks with an aggregate bandwidth of 100 Gbps. Over one week of operation, our deployment served more than 50,000 real users. The experience demonstrates that TapDance can be practically realized at ISP scale with good performance and at a reasonable cost, potentially paving the way for long-term, large-scale deployments of TapDance or other refraction networking schemes in the future.

Over the past 20 years, websites have grown increasingly complex and interconnected. In 2016, only a negligible number of sites are dependency free, and over 90% of sites rely on external content. In this paper, we investigate the current state of web dependencies and explore two security challenges associated with the increasing reliance on external services: (1) the expanded attack surface associated with serving unknown, implicitly trusted third-party content, and (2) how the increased set of external dependencies impacts HTTPS adoption. We hope that by shedding light on these issues, we can encourage developers to consider the security risks associated with serving third-party content and prompt service providers to more widely deploy HTTPS.

As HTTPS deployment grows, middlebox and antivirus products are increasingly intercepting TLS connections to retain visibility into network traffic. In this work, we present a comprehensive study on the prevalence and impact of HTTPS interception. First, we show that web servers can detect interception by identifying a mismatch between the HTTP User-Agent header and TLS client behavior. We characterize the TLS handshakes of major browsers and popular interception products, which we use to build a set of heuristics to detect interception and identify the responsible product. We deploy these heuristics at three large network providers: (1) Mozilla Firefox update servers, (2) a set of popular e-commerce sites, and (3) the Cloudflare content distribution network. We find more than an order of magnitude more interception than previously estimated and with dramatic impact on connection security. To understand why security suffers, we investigate popular middleboxes and client- side security software, finding that nearly all reduce connection security and many introduce severe vulnerabilities. Drawing on our measurements, we conclude with a discussion on recent proposals to safely monitor HTTPS and recommendations for the security community.

Several recent standards, including NIST SP 800-56A and RFC 5114, advocate the use of “DSA” parameters for Diffie-Hellman key exchange. While it is possible to use such parameters securely, additional validation checks are necessary to prevent well known and potentially devastating attacks. In this paper, we observe that many Diffie-Hellman implementations do not properly validate key exchange inputs. Combined with other protocol properties and implementation choices, this can radically decrease security. We measure the prevalence of these parameter choices in the wild for HTTPS, POP3S, SMTP with STARTTLS, SSH, IKEv1, and IKEv2, finding millions of hosts using DSA and other non-“safe” primes for Diffie-Hellman key exchange, many of them in combination with potentially vulnerable behaviors. We examine over 20 open-source cryptographic libraries and applications and observe that until January 2016, not a single one validated subgroup orders by default. We found feasible full or partial key recovery vulnerabilities in OpenSSL, the Exim mail server, the Unbound DNS client, and Amazon’s load balancer, as well as susceptibility to weaker attacks in many other applications.

Industrial control systems have become ubiquitous, enabling the remote, electronic control of physical equipment and sensors. Originally designed to operate on closed networks, the protocols used by these devices have no built-in security. However, despite this, an alarming number of systems are connected to the public Internet and an attacker who finds a device often can cause catastrophic damage to physical infrastructure. We consider two aspects of ICS security in this work: (1) what devices have been inadvertently exposed on the public Internet, and (2) who is searching for vulnerable systems. First, we implement five common SCADA protocols in ZMap and conduct a survey of the public IPv4 address space finding more than 60K publicly accessible systems. Second, we use a large network telescope and high-interaction honeypots to find and profile actors searching for devices. We hope that our findings can both motivate and inform future work on securing industrial control systems.

In this paper we explore the notion of a secure kiosk, a trusted computing platform built using off-the-shelf components. We demonstrate how kiosks serve as convenient primitives when designing secure computing protocols, as they allow for a very prescribed set of assumptions to be made about a system. We begin by defining the necessary properties of a kiosk, and then explain how each of these properties can (or cannot) be attained using current off-the-shelf hardware and software components. We construct a proof-of-concept implementation using TPM hardware and Windows 10. We also provide ASKVote, the Attestable and Secure Voting protocol to demonstrate the flexibility gained from the use of kiosks in a larger secure system.

TLS has the potential to provide strong protection against network-based attackers and mass surveillance, but many implementations take security shortcuts in order to reduce the costs of cryptographic computations and network round trips. We report the results of a nine-week study that measures the use and security impact of these shortcuts for HTTPS sites among Alexa Top Million domains. We find widespread deployment of DHE and ECDHE private value reuse, TLS session resumption, and TLS session tickets. These practices greatly reduce the protection afforded by forward secrecy: connections to 38% of Top Million HTTPS sites are vulnerable to decryption if the server is compromised up to 24 hours later, and 10% up to 30 days later, regardless of the selected cipher suite. We also investigate the practice of TLS secrets and session state being shared across domains, finding that in some cases, the theft of a single secret value can compromise connections to tens of thousands of sites. These results suggest that site operators need to better understand the tradeoffs between optimizing TLS performance and providing strong security, particularly when faced with nation-state attackers with a history of aggressive, large-scale surveillance.

The HTTPS certificate ecosystem has been of great interest to the measurement and security communities. Without any ground truth, researchers have attempted to study this PKI from a variety of fragmented perspectives, including passively monitored networks, scans of the popular domains or the IPv4 address space, search engines such as Censys, and Certificate Transparency (CT) logs. In this work, we comparatively analyze all these perspectives. We find that aggregated CT logs and Censys snapshots have many properties that complement each other, and that together they encompass over 99% of all certificates found by any of these techniques. However, they still miss 1.5% of certificates observed in a crawl of all domains in .com, .net, and .org. We go on to illustrate how this combined perspective affects results from previous studies. In light of these findings, we have worked with the operators of Censys to incorporate CT log data into its results going forward, and we recommend that future HTTPS measurement adopt this new vantage.

The World Wide Web has become the most common platform for building applications and delivering content. Yet despite years of research, the web continues to face severe security challenges related to data integrity and confidentiality. Rather than continuing the exploit-and-patch cycle, we propose addressing these challenges at an architectural level, by supplementing the web’s existing connection-based and server-based security models with a new approach: content-based security. With this approach, content is directly signed and encrypted at rest, enabling it to be delivered via any path and then validated by the browser. We explore how this new architectural approach can be applied to the web and analyze its security benefits. We then discuss a broad research agenda to realize this vision and the challenges that must be overcome.

We present DROWN, a novel cross-protocol attack on TLS that uses a server supporting SSLv2 as an oracle to decrypt modern TLS connections.

We introduce two versions of the attack. The more general form exploits multiple unnoticed protocol flaws in SSLv2 to develop a new and stronger variant of the Bleichenbacher RSA padding-oracle attack. To decrypt a 2048-bit RSA TLS ciphertext, an attacker must observe 1,000 TLS handshakes, initiate 40,000 SSLv2 connections, and perform 2⁵⁰ offline work. The victim client never initiates SSLv2 connections. We implemented the attack and can decrypt a TLS 1.2 handshake using 2048-bit RSA in under 8 hours, at a cost of $440 on Amazon EC2. Using Internet-wide scans, we find that 33% of all HTTPS servers and 22% of those with browser-trusted certificates are vulnerable to this protocol-level attack due to widespread key and certificate reuse.

For an even cheaper attack, we apply our new techniques together with a newly discovered vulnerability in OpenSSL that was present in releases from 1998 to early 2015. Given an unpatched SSLv2 server to use as an oracle, we can decrypt a TLS ciphertext in one minute on a single CPU—fast enough to enable man-in-the-middle attacks against modern browsers. We find that 26% of HTTPS servers are vulnerable to this attack.

We further observe that the QUIC protocol is vulnerable to a variant of our attack that allows an attacker to impersonate a server indefinitely after performing as few as 2¹⁷ SSLv2 connections and 2⁵⁸ offline work.

We conclude that SSLv2 is not only weak, but actively harmful to the TLS ecosystem.

Once pervasive, the File Transfer Protocol (FTP) has been largely supplanted by HTTP, SCP, and BitTorrent for transferring data between hosts. Yet, in a comprehensive analysis of the FTP ecosystem as of 2015, we find that there are still more than 13 million FTP servers in the IPv4 address space, 1.1 million of which allow “anonymous” (public) access. These anonymous FTP servers leak sensitive information, such as tax documents and cryptographic secrets. More than 20,000 FTP servers allow public write access, which has facilitated malicious actors’ use of free storage as well as malware deployment and click-fraud attacks. We further investigate real-world attacks by deploying eight FTP honeypots, shedding light on how attackers are abusing and exploiting vulnerable servers. We conclude with lessons and recommendations for securing FTP.

App-based deception attacks are increasingly a problem on mobile devices and they are used to steal passwords, credit card numbers, text messages, etc. Current versions of Android are susceptible to these attacks. Recently, Bianchi et al. proposed a novel solution (“What the App is That”) that included a host-based system to identify apps to users via a security indicator and help assure them that their input goes to the identified apps. Unfortunately, we found that the solution has a significant side channel vulnerability as well as susceptibility to clickjacking that allow non-privileged malware to completely compromise the defenses, and successfully steal passwords or other keyboard input. We discuss the vulnerabilities found, propose possible defenses, and then evaluate the defenses against different types of UI deception attacks.

We investigate the security of Diffie-Hellman key exchange as used in popular Internet protocols and find it to be less secure than widely believed. First, we present Logjam, a novel flaw in TLS that lets a man-in-the-middle downgrade connections to “export-grade” Diffie-Hellman. To carry out this attack, we implement the number field sieve discrete log algorithm. After a week-long precomputation for a specified 512-bit group, we can compute arbitrary discrete logs in that group in about a minute. We find that 82% of vulnerable servers use a single 512-bit group, allowing us to compromise connections to 7% of Alexa Top Million HTTPS sites. In response, major browsers are being changed to reject short groups.

We go on to consider Diffie-Hellman with 768- and 1024-bit groups. We estimate that even in the 1024-bit case, the computations are plausible given nation-state resources. A small number of fixed or standardized groups are used by millions of servers; performing precomputation for a single 1024-bit group would allow passive eavesdropping on 18% of popular HTTPS sites, and a second group would allow decryption of traffic to 66% of IPsec VPNs and 26% of SSH servers. A close reading of published NSA leaks shows that the agency’s attacks on VPNs are consistent with having achieved such a break. We conclude that moving to stronger key exchange methods should be a priority for the Internet community.

Fast Internet-wide scanning has opened new avenues for security research, ranging from uncovering widespread vulnerabilities in random number generators to tracking the evolving impact of Heartbleed. However, this technique still requires significant effort: even simple questions, such as, “What models of embedded devices prefer CBC ciphers?”, require developing an application scanner, manually identifying and tagging devices, negotiating with network administrators, and responding to abuse complaints. In this paper, we introduce Censys, a public search engine and data processing facility backed by data collected from ongoing Internet-wide scans. Designed to help researchers answer security-related questions, Censys supports full-text searches on protocol banners and querying a wide range of derived fields (e.g., 443.https.cipher). It can identify specific vulnerable devices and networks and generate statistical reports on broad usage patterns and trends. Censys returns these results in sub-second time, dramatically reducing the effort of understanding the hosts that comprise the Internet. We present the search engine architecture and experimentally evaluate its performance. We also explore Censys’s applications and show how questions asked in recent studies become simple to answer.

The SMTP protocol is responsible for carrying some of users’ most intimate communication, but like other Internet protocols, authentication and confidentiality were added only as an afterthought. In this work, we present the first report on global adoption rates of SMTP security extensions, including StartTLS, SPF, DKIM, and DMARC. We present data from two perspectives: SMTP server configurations for the Alexa Top Million domains, and over a year of SMTP connections to and from Gmail. We find that the top mail providers (e.g., Gmail, Yahoo, Outlook) all proactively encrypt and authenticate messages. However, these best practices have yet to reach widespread adoption in a long tail of over 700,000 SMTP servers, of which only 35% successfully configure encryption and 1.1% specify a DMARC authentication policy. This security patchwork—paired with SMTP policies that favor failing open to allow gradual deployment—exposes users to attackers who downgrade TLS connections in favor of cleartext and who falsify MX records to reroute messages. We present evidence of such attacks in the wild, highlighting seven countries where more than 20% of inbound Gmail messages arrive in cleartext due to network attackers.

In the world’s largest-ever deployment of online voting, the iVote Internet voting system was trusted for the return of 280,000 ballots in the 2015 state election in New South Wales, Australia. During the election, we performed an independent security analysis of parts of the live iVote system and uncovered severe vulnerabilities that could be leveraged to manipulate votes, violate ballot privacy, and subvert the verification mechanism. These vulnerabilities do not seem to have been detected by the election authorities before we disclosed them, despite a pre-election security review and despite the system having run in a live state election for five days. One vulnerability, the result of including analytics software from an insecure external server, exposed some votes to complete compromise of privacy and integrity. At least one parliamentary seat was decided by a margin much smaller than the number of votes taken while the system was vulnerable. We also found protocol flaws, including vote verification that was itself susceptible to manipulation. This incident underscores the difficulty of conducting secure elections online and carries lessons for voters, election officials, and the e-voting research community.

Embedded devices with web interfaces are prevalent, but, due to memory and processing constraints, implementations typically make use of Common Gateway Interface (CGI) binaries written in low-level, memory-unsafe languages. This creates the possibility of memory corruption attacks as well as traditional web attacks. We present Umbra, an application-layer firewall specifically designed for protecting web interfaces in embedded devices. By acting as a “friendly man-in-the-middle,” Umbra can protect against attacks such as cross-site request forgery (CSRF), information leaks, and authentication bypass vulnerabilities. We evaluate Umbra’s security by analyzing recent vulnerabilities listed in the CVE database from several embedded device vendors and find that it would have prevented half of these vulnerabilities. We also show that Umbra comfortably runs within the constraints of an embedded system while incurring minimal performance overhead.

Several attacks against physical pin-tumbler locks require access to one or more key blanks to perform. These attacks include bumping, impressioning, rights amplification, and teleduplication. To mitigate these attacks, many lock systems rely on restricted keyways and use blanks that are not sold to the general public, making it harder for attackers to obtain them. Often the key blank designs themselves are patented, further discouraging distribution or manufacture by even skilled machinists. In this paper, we investigate the impact that emerging rapid-prototyping—or 3D-printing—tools have on the security of these restricted keyway systems. We find that commodity 3D printers are able to produce key blanks and pre-cut keys with enough resolution to work in several commonly used pin-tumbler locks and that their material is strong enough to withstand the requirements to perform the aforementioned attacks. In addition, in order to demonstrate the low skill requirements necessary to perform these attacks, we develop a tool that automatically generates a 3D-printable CAD model of a key blank using only a single picture of a lock’s keyway. This tool allows us to rapidly manufacture key blanks for restricted keyways that were previously difficult to make or buy. Finally, we discuss possible mitigations for these attacks that lock manufacturers, installers, and users can perform to protect their assets.

The Heartbleed vulnerability took the Internet by surprise in April 2014. The vulnerability, one of the most consequential since the advent of the commercial Internet, allowed attackers to remotely read protected memory from an estimated 24–55% of popular HTTPS sites. In this work, we perform a comprehensive, measurement-based analysis of the vulnerability’s impact, including (1) tracking the vulnerable population, (2) monitoring patching behavior over time, (3) assessing the impact on the HTTPS certificate ecosystem, and (4) exposing real attacks that attempted to exploit the bug. Furthermore, we conduct a large-scale vulnerability notification experiment involving 150,000 hosts and observe a nearly 50% increase in patching by notified hosts. Drawing upon these analyses, we discuss what went well and what went poorly, in an effort to understand how the technical community can respond more effectively to such events in the future.

Estonia was the first country in the world to use Internet voting nationally, and today more than 30% of its ballots are cast online. In this paper, we analyze the security of the Estonian I-voting system based on a combination of in-person election observation, code review, and adversarial testing. Adopting a threat model that considers the advanced threats faced by a national election system—including dishonest insiders and state-sponsored attacks—we find that the I-voting system has serious architectural limitations and procedural gaps that potentially jeopardize the integrity of elections. In experimental attacks on a reproduction of the system, we demonstrate how such attackers could target the election servers or voters’ clients to alter election results or undermine the legitimacy of the system. Our findings illustrate the practical obstacles to Internet voting in the modern world, and they carry lessons for Estonia, for other countries considering adopting such systems, and for the security research community.

In a multi-level election, voters are divided into groups, an election is held within each group, and some deterministic procedure is used to combine the group results to determine the overall election result. Examples of multi-level elections include U.S. presidential elections and some parliamentary elections (such as those with regional groupings of voters). The results of such an election can hinge on a few votes in one group, while being insensitive to large shifts within other groups. These disparities create opportunities to focus election integrity efforts in the places where they have the highest leverage. We consider how to improve the efficiency of post-election audits, such as those that compare paper ballots to corresponding electronic records, in multi-level elections. We evaluate our proposed solutions using data from past elections.

Advanced imaging technologies are a new class of people screening systems used at airports and other sensitive environments to detect metallic as well as nonmetallic contraband. We present the first independent security evaluation of such a system, the Rapiscan Secure 1000 full-body scanner, which was widely deployed at airport checkpoints in the U.S. from 2009 until 2013. We find that the system provides weak protection against adaptive adversaries: It is possible to conceal knives, guns, and explosives from detection by exploiting properties of the device’s backscatter X-ray technology. We also investigate cyberphysical threats and propose novel attacks that use malicious software and hardware to compromise the the effectiveness, safety, and privacy of the device. Overall, our findings paint a mixed picture of the Secure 1000 that carries lessons for the design, evaluation, and operation of advanced imaging technologies, for the ongoing public debate concerning their use, and for cyberphysical security more broadly.

In response to increasingly sophisticated state-sponsored Internet censorship, recent work has proposed a new approach to censorship resistance: end-to-middle proxying. This concept, developed in systems such as Telex, Decoy Routing, and Cirripede, moves anticensorship technology into the core of the network, at large ISPs outside the censoring country. In this paper, we focus on two technical obstacles to the deployment of certain end-to-middle schemes: the need to selectively block flows and the need to observe both directions of a connection. We propose a new construction, TapDance, that removes these requirements. TapDance employs a novel TCP-level technique that allows the anticensorship station at an ISP to function as a passive network tap, without an inline blocking component. We also apply a novel steganographic encoding to embed control messages in TLS ciphertext, allowing us to operate on HTTPS connections even under asymmetric routing. We implement and evaluate a TapDance prototype that demonstrates how the system could function with minimal impact on an ISP’s network operations.

While it is widely known that port scanning is widespread, neither the scanning landscape nor the defensive reactions of network operators have been measured at Internet scale. In this work, we analyze data from a large network telescope to study scanning activity from the past year, uncovering large horizontal scan operations and identifying broad patterns in scanning behavior. We present an analysis of who is scanning, what services are being targeted, and the impact of new scanners on the overall landscape. We also analyze the scanning behavior triggered by recent vulnerabilities in Linksys routers, OpenSSL, and NTP. We empirically analyze the defensive behaviors that organizations employ against scanning, shedding light on who detects scanning behavior, which networks blacklist scanning, and how scan recipients respond to scans conducted by researchers. We conclude with recommendations for institutions performing scans and with implications of recent changes in scanning behavior for researchers and network operators.

The safety critical nature of traffic infrastructure requires that it be secure against computer-based attacks, but this is not always the case. We investigate a networked traffic signal system currently deployed in the United States and discover a number of security flaws that exist due to systemic failures by the designers. We leverage these flaws to create attacks which gain control of the system, and we successfully demonstrate them on the deployment in coordination with authorities. Our attacks show that an adversary can control traffic infrastructure to cause disruption, degrade safety, or gain an unfair advantage. We make recommendations on how to improve existing systems and discuss the lessons learned for embedded systems security in general.

We introduce optimizations to the ZMap network scanner that achieve a 10-fold increase in maximum scan rate. By parallelizing address generation, introducing an improved blacklisting algorithm, and using zero-copy NIC access, we drive ZMap to nearly the maximum throughput of 10 gigabit Ethernet, almost 15 million probes per second. With these changes, ZMap can comprehensively scan for a single TCP port across the entire public IPv4 address space in 4.5 minutes given adequate upstream bandwidth. We consider the implications of such rapid scanning for both defenders and attackers, and we briefly discuss a range of potential applications.

In this paper, we perform a review of elliptic curve cryptography (ECC), as it is used in practice today, in order to reveal unique mistakes and vulnerabilities that arise in implementations of ECC. We study four popular protocols that make use of this type of public-key cryptography: Bitcoin, secure shell (SSH), transport layer security (TLS), and the Austrian e-ID card. We are pleased to observe that about 1 in 10 systems support ECC across the TLS and SSH protocols. However, we find that despite the high stakes of money, access and resources protected by ECC, implementations suffer from vulnerabilities similar to those that plague previous cryptographic systems.

Many companies have recently started to offer wearable computing devices including glasses, bracelets, and watches. While this technology enables exciting new applications, it also poses new security and privacy concerns. In this work, we explore these implications and analyze the impact of one of the first networked wearable devices—smartwatches—on an academic environment. As a proof of concept, we develop an application for the Pebble smartwatch called ConTest that would allow dishonest students to inconspicuously collaborate on multiple-choice exams in real time, using a cloud-based service, a smartphone, and a client application on the watch. We discuss the broader implications of this technology, suggest hardware and software approaches that can be used to prevent such attacks, and pose questions for future research.

We report the results of a large-scale measurement study of the HTTPS certificate ecosystem—the public-key infrastructure that underlies nearly all secure web communications. Using data collected by performing 110 Internet-wide scans over 14 months, we gain detailed and temporally fine-grained visibility into this otherwise opaque area of security-critical infrastructure. We investigate the trust relationships among root authorities, intermediate authorities, and the leaf certificates used by web servers, ultimately identifying and classifying more than 1,800 entities that are able to issue certificates vouching for the identity of any website. We uncover practices that may put the security of the ecosystem at risk, and we identify frequent configuration problems that lead to user-facing errors and potential vulnerabilities. We conclude with lessons and recommendations to ensure the long-term health and security of the certificate ecosystem.

Internet-wide network scanning has numerous security applications, including exposing new vulnerabilities and tracking the adoption of defensive mechanisms, but probing the entire public address space with existing tools is both difficult and slow. We introduce ZMap, a modular, open-source network scanner specifically architected to perform Internet-wide scans and capable of surveying the entire IPv4 address space in under 45 minutes from user space on a single machine, approaching the theoretical maximum speed of gigabit Ethernet. We present the scanner architecture, experimentally characterize its performance and accuracy, and explore the security implications of high speed Internet-scale network surveys, both offensive and defensive. We also discuss best practices for good Internet citizenship when performing Internet-wide surveys, informed by our own experiences conducting a long-term research survey over the past year.

Out-of-band, lights-out management has become a standard feature on many servers, but while this technology can be a boon for system administrators, it also presents a new and interesting vector for attack. This paper examines the security implications of the Intelligent Platform Management Interface (IPMI), which is implemented on server motherboards using an embedded Baseboard Management Controller (BMC). We consider the threats posed by an incorrectly implemented IPMI and present evidence that IPMI vulnerabilities may be widespread. We analyze a major OEM’s IPMI implementation and discover that it is riddled with textbook vulnerabilities, some of which would allow a remote attacker to gain root access to the BMC and potentially take control of the host system. Using data from Internet-wide scans, we find that there are at least 100,000 IPMI-enabled servers (across three large vendors) running on publicly accessible IP addresses, contrary to recommended best practice. Finally, we suggest defensive strategies for servers currently deployed and propose avenues for future work.

The Iranian government operates one of the largest and most sophisticated Internet censorship regimes in the world, but the mechanisms it employs have received little research attention, primarily due to lack of access to network connections within the country and personal risks to Iranian citizens who take part. In this paper, we examine the status of Internet censorship in Iran based on network measurements conducted from a major Iranian ISP during the lead up to the June 2013 presidential election. We measure the scope of the censorship by probing Alexa’s top 500 websites in 18 different categories. We investigate the technical mechanisms used for HTTP Host–based blocking, keyword filtering, DNS hijacking, and protocol-based throttling. Finally, we map the network topology of the censorship infrastructure and find evidence that it relies heavily on centralized equipment, a property that might be fruitfully exploited by next generation approaches to censorship circumvention.

The existing HTTPS public-key infrastructure (PKI) uses a coarse-grained trust model: either a certificate authority (CA) is trusted by browsers to vouch for the identity of any domain or it is not trusted at all. More than a thousand root and intermediate CAs can currently sign certificates for any domain and be trusted by popular browsers. This violates the principle of least privilege and creates an excessively large attack surface, as highlighted by recent CA compromises. In this paper, we present CAge, a mechanism that browser makers can apply to drastically reduce the excessive trust placed in CAs without fundamentally altering the CA ecosystem or breaking existing practice. CAge works by imposing restrictions on the set of top-level domains (TLDs) for which each CA is trusted to sign. Our key observation, based on an Internet-wide survey of TLS certificates, is that CAs commonly sign for only a handful of TLDs; in fact, 90% of CAs have signed certificates for domains in fewer than ten TLDs, and only 35% have ever signed a certificate for a domain in .com. We show that it is possible to algorithmically infer reasonable restrictions on CAs’ trusted scopes based on this behavior, and we present evidence that browser-enforced inferred scopes would be a durable and effective way to reduce the attack surface of the HTTPS PKI. We find that simple inference rules can reduce the attack surface by nearly a factor of ten without hindering 99% of CA signing activity over a six-month period.

RSA and DSA can fail catastrophically when used with malfunctioning random number generators, but the extent to which these problems arise in practice has never been comprehensively studied at Internet scale. We perform the largest ever network survey of TLS and SSH servers and present evidence that vulnerable keys are surprisingly widespread. We find that 0.75% of TLS certificates share keys due to insufficient entropy during key generation, and we suspect that another 1.70% come from the same faulty implementations and may be susceptible to compromise. Even more alarmingly, we are able to obtain RSA private keys for 0.50% of TLS hosts and 0.03% of SSH hosts, because their public keys shared nontrivial common factors due to entropy problems, and DSA private keys for 1.03% of SSH hosts, because of insufficient signature randomness. We cluster and investigate the vulnerable hosts, finding that the vast majority appear to be headless or embedded devices. In experiments with three software components commonly used by these devices, we are able to reproduce the vulnerabilities and identify specific software behaviors that induce them, including a boot-time entropy hole in the Linux random number generator. Finally, we suggest defenses and draw lessons for developers, users, and the security community.

In 2010, Washington, D.C. developed an Internet voting pilot project that was intended to allow overseas absentee voters to cast their ballots using a website. Prior to deploying the system in the general election, the District held a unique public trial: a mock election during which anyone was invited to test the system or attempt to compromise its security. This paper describes our experience participating in this trial. Within 48 hours of the system going live, we had gained near-complete control of the election server. We successfully changed every vote and revealed almost every secret ballot. Election officials did not detect our intrusion for nearly two business days—and might have remained unaware for far longer had we not deliberately left a prominent clue. This case study—the first (to our knowledge) to analyze the security of a government Internet voting system from the perspective of an attacker in a realistic pre-election deployment—attempts to illuminate the practical challenges of securing online voting as practiced today by a growing number of jurisdictions.

In this paper, we present Telex, a new approach to resisting state-level Internet censorship. Rather than attempting to win the cat-and-mouse game of finding open proxies, we leverage censors’ unwillingness to completely block day-to-day Internet access. In effect, Telex converts innocuous, unblocked websites into proxies, without their explicit collaboration. We envision that friendly ISPs would deploy Telex stations on paths between censors’ networks and popular, uncensored Internet destinations. Telex stations would monitor seemingly innocuous flows for a special “tag” and transparently divert them to a forbidden website or service instead. We propose a new cryptographic scheme based on elliptic curves for tagging TLS handshakes such that the tag is visible to a Telex station but not to a censor. In addition, we use our tagging scheme to build a protocol that allows clients to connect to Telex stations while resisting both passive and active attacks. We also present a proof-of-concept implementation that demonstrates the feasibility of our system.

China filters Internet traffic in and out of the country. In order to circumvent the firewall, it is helpful to know where the filtering occurs. In this work, we explore the AS-level topology of China’s network, and probe the firewall to find the locations of filtering devices. We find that even though most filtering occurs in border ASes, choke points also exist in many provincial networks. The result suggests that two major ISPs in China have different approaches for placing filtering devices.

Many IT departments use remote administration products to configure, monitor, and maintain the systems they manage. These tools can be beneficial in the right hands, but they can also be devastating if attackers exploit them to seize control of machines. As a case study, we analyze the security of a remote administration product called Absolute Manage. We find that the system’s communication protocol suffers from serious design flaws and fails to provide adequate integrity, confidentiality, or authentication. Attackers can exploit these vulnerabilities to issue unauthorized commands on client systems and execute arbitrary code with administrator privileges. These blatant vulnerabilities suggest that remote administration tools require increased scrutiny from the security community. We recommend that developers adopt defensive designs that limit the damage attackers can cause if they gain control.

Research about weaknesses in deployed electronic voting systems raises a variety of pressing ethical concerns. In addition to ethical issues common to vulnerability research, such as the potential harms and benefits of vulnerability disclosure, electronic voting researchers face questions that flow from the unique and important role voting plays in modern democratic societies. Should researchers worry that their own work (not unlike the flaws they study) could sway an election outcome? When elected officials authorize a security review, how should researchers address the conflicted interests of these incumbent politicians, who may have powerful incentives to downplay problems, and might in principle be in a position to exploit knowledge about vulnerabilities when they stand for re-election? How should researchers address the risk that identifying specific flaws will lead to a false sense of security, after those particular problems have been resolved? This paper makes an early effort to address these and other questions with reference to experience from previous e-voting security reviews. We hope our provisional analysis will help practicing researchers anticipate and address ethical issues in future studies.

Elections in India are conducted almost exclusively using electronic voting machines developed over the past two decades by a pair of government-owned companies. These devices, known in India as EVMs, have been praised for their simple design, ease of use, and reliability, but recently they have also been criticized following widespread reports of election irregularities. Despite this criticism, many details of the machines’ design have never been publicly disclosed, and they have not been subjected to a rigorous, independent security evaluation. In this paper, we present a security analysis of a real Indian EVM obtained from an anonymous source. We describe the machine’s design and operation in detail, and we evaluate its security in light of relevant election procedures. We conclude that in spite of the machines’ simplicity and minimal software trusted computing base, they are vulnerable to serious attacks that can alter election results and violate the secrecy of the ballot. We demonstrate two attacks, implemented using custom hardware, which could be carried out by dishonest election insiders or other criminals with only brief physical access to the machines. This case study carries important lessons for Indian elections and for electronic voting security more generally.

This paper presents two kinds of attacks based on crawling the DHTs used for distributed BitTorrent tracking. First, we show how pirates can use crawling to rebuild BitTorrent search engines just a few hours after they are shut down (crawling for fun). Second, we show how content owners can use related techniques to monitor pirates’ behavior in preparation for legal attacks and negate any perceived anonymity of the decentralized BitTorrent architecture (crawling for profit).

We validate these attacks and measure their performance with a crawler we developed for the Vuze DHT. We find that we can establish a search engine with over one million torrents in under two hours using a single desktop PC. We also track 7.9 million IP addresses downloading 1.5 million torrents over 16 days. These results imply that shifting from centralized BitTorrent tracking to DHT-based tracking will have mixed results for the file sharing arms race. While it will likely make illicit torrents harder to quash, it will not help users hide their activities.

This paper introduces a captcha based on upright orientation of line drawings rendered from 3D models. The models are selected from a large database, and images are rendered from random viewpoints, affording many different drawings from a single 3D model. The captcha presents the user with a set of images, and the user must choose an upright orientation for each image. This task generally requires understanding of the semantic content of the image, which is believed to be difficult for automatic algorithms. We describe a process called covert filtering whereby the image database can be continually refreshed with drawings that are known to have a high success rate for humans, by inserting randomly into the captcha new images to be evaluated. Our analysis shows that covert filtering can ensure that captchas are likely to be solvable by humans while deterring attackers who wish to learn a portion of the database. We performed several user studies that evaluate how effectively people can solve the captcha. Comparing these results to an attack based on machine learning, we find that humans possess a substantial performance advantage over computers.

Researchers at the University of Washington recently proposed Vanish, a system for creating messages that automatically “self-destruct” after a period of time. Vanish works by encrypting each message with a random key and storing shares of the key in a large, public distributed hash table (DHT). Normally, DHTs expunge data older than a certain age. After they expire, the key is permanently lost, and the encrypted data is permanently unreadable. Vanish is an interesting approach to an important privacy problem, but, in its current form, it is insecure. In this paper, we defeat the deployed Vanish implementation, explain how the original paper’s security analysis is flawed, and draw lessons for future system designs.

We present two Sybil attacks against the current Vanish implementation, which stores its encryption keys in the million-node Vuze BitTorrent DHT. These attacks work by continuously crawling the DHT and saving each stored value before it ages out. They can efficiently recover keys for more than 99% of Vanish messages. We show that the dominant cost of these attacks is network data transfer, not memory usage as the Vanish authors expected, and that the total cost is two orders of magnitude less than they estimated. While we consider potential defenses, we conclude that public DHTs like Vuze probably cannot provide strong security for Vanish.

A secure voting machine design must withstand new attacks devised throughout its multidecade service lifetime. In this paper, we give a case study of the longterm security of a voting machine, the Sequoia AVC Advantage, whose design dates back to the early 80s. The AVC Advantage was designed with promising security features: its software is stored entirely in read-only memory and the hardware refuses to execute instructions fetched from RAM. Nevertheless, we demonstrate that an attacker can induce the AVC Advantage to misbehave in arbitrary ways—including changing the outcome of an election—by means of a memory cartridge containing a specially-formatted payload. Our attack makes essential use of a recently-invented exploitation technique called return-oriented programming, adapted here to the Z80 processor. In return-oriented programming, short snippets of benign code already present in the system are combined to yield malicious behavior. Our results demonstrate the relevance of recent ideas from systems security to voting machine research, and vice versa. We had no access either to source code or documentation beyond that available on Sequoia’s web site. We have created a complete vote-stealing demonstration exploit and verified that it works correctly on the actual hardware.

This paper presents a novel technique for authenticating physical documents based on random, naturally occurring imperfections in paper texture. We introduce a new method for measuring the three-dimensional surface of a page using only a commodity scanner and without modifying the document in any way. From this physical feature, we generate a concise fingerprint that uniquely identifies the document. Our technique is secure against counterfeiting and robust to harsh handling; it can be used even before any content is printed on a page. It has a wide range of applications, including detecting forged currency and tickets, authenticating passports, and halting counterfeit goods. Document identification could also be applied maliciously to de-anonymize printed surveys and to compromise the secrecy of paper ballots.

Contrary to popular assumption, DRAMs used in most modern computers retain their contents for seconds to minutes after power is lost, even at operating temperatures and even if removed from a motherboard. Although DRAMs become less reliable when they are not refreshed, they are not immediately erased, and their contents persist sufficiently for malicious (or forensic) acquisition of usable full-system memory images. We show that this phenomenon limits the ability of an operating system to protect cryptographic key material from an attacker with physical access. We use cold reboots to mount attacks on popular disk encryption systems — BitLocker, FileVault, dm-crypt, and TrueCrypt — using no special devices or materials. We experimentally characterize the extent and predictability of memory remanence and report that remanence times can be increased dramatically with simple techniques. We offer new algorithms for finding cryptographic keys in memory images and for correcting errors caused by bit decay. Though we discuss several strategies for partially mitigating these risks, we know of no simple remedy that would eliminate them.

Generation of random numbers is a critical component of existing post-election auditing techniques. Recent work has largely discouraged the use of all pseudorandom number generators, including cryptographically secure pseudorandom number generators (CSPRNGs), for this purpose, instead recommending the sole use of observable physical techniques. In particular, simple dice rolling has received a great deal of positive attention. The typical justification for this recommendation is that those less comfortable with mathematics prefer a simple, observable technique. This paper takes a contrary view. Simple, observable techniques like dice rolling are not necessarily robust against sleight of hand and other forms of fraud, and attempts to harden them against fraud can dramatically increase their complexity. With simple dice rolling, we know of no techniques that provide citizens with a reasonable means of verifying that fraud did not occur during the roll process. CSPRNGs, used properly, can be simple, robust, and verifiable, and they allow for the use of auditing techniques that might otherwise be impractical. While we understand initial skepticism towards this option, we argue that appropriate use of CSPRNGs would strengthen audit security.

In light of the systemic vulnerabilities uncovered by recent reviews of deployed e-voting systems, the surest way to secure the voting process would be to scrap the existing systems and design new ones. Unfortunately, engineering new systems will take years, and many jurisdictions are unlikely to be able to afford new equipment in the near future. In this paper we ask how jurisdictions can make the best use of the equipment they already own until they can replace it. Starting from current practice, we propose defenses that involve new but realistic procedures, modest changes to existing software, and no changes to existing hardware. Our techniques achieve greatly improved protection against outsider attacks: they provide containment of viral spread, improve the integrity of vote tabulation, and offer some detection of individual compromised devices. They do not provide security against insiders with access to election management systems, which appears to require significantly greater changes to the existing systems.

Several important security protocols require parties to perform computations based on random challenges. Traditionally, proving that the challenges were randomly chosen has required interactive communication among the parties or the existence of a trusted server. We offer an alternative solution where challenges are harvested from oblivious servers on the Internet. This paper describes a framework for deriving “harvested challenges” by mixing data from various pre-existing online sources. While individual sources may become predictable or fall under adversarial control, we provide a policy language that allows application developers to specify combinations of sources that meet their security needs. Participants can then convince each other that their challenges were formed freshly and in accordance with the policy. We present Combine, an open source implementation of our framework, and show how it can be applied to a variety of applications, including remote storage auditing and non-interactive client puzzles.

Election audit procedures usually rely on precinct based recounts, in which workers manually review all paper ballots from selected polling places, but these recounts can be expensive due to the labor required. This paper proposes an alternative audit strategy that allows machines to perform most of the work. Precincts are recounted using recounting machines, and their output is manually audited using efficient ballot sampling techniques. This strategy can achieve equal or greater confidence than precinct-based auditing at a significantly lower cost while protecting voter privacy better than previous ballot-based auditing methods. We show how to determine which ballots to audit against the recounting machines’ records and compare this new approach to precinct-based audits in the context of Virginia’s November 2006 election. Far fewer ballots need to be audited by hand using our approach. We also explore extensions to these techniques, such as varying individual ballots’ audit probabilities based on the votes they contain, that promise further efficiency gains.

This paper presents a fully independent security study of a Diebold AccuVote-TS voting machine, including its hardware and software. We obtained the machine from a private party. Analysis of the machine, in light of real election procedures, shows that it is vulnerable to extremely serious attacks. For example, an attacker who gets physical access to a machine or its removable memory card for as little as one minute could install malicious code; malicious code on a machine could steal votes undetectably, modifying all records, logs, and counters to be consistent with the fraudulent vote count it creates. An attacker could also create malicious code that spreads automatically and silently from machine to machine during normal election activities — a voting-machine virus. We have constructed working demonstrations of these attacks in our lab. Mitigating these threats will require changes to the voting machine’s hardware and software and the adoption of more rigorous election procedures.

In the fall of 2005, problems discovered in two Sony-BMG compact disc copy protection systems, XCP and MediaMax, triggered a public uproar that ultimately led to class-action litigation and the recall of millions of discs. We present an in-depth analysis of these technologies, including their design, implementation, and deployment. The systems are surprisingly complex and suffer from a diverse array of flaws that weaken their content protection and expose users to serious security and privacy risks. Their complexity, and their failure, makes them an interesting case study of digital rights management that carries valuable lessons for content companies, DRM vendors, policymakers, end users, and the security community.

Computer users are asked to generate, keep secret, and recall an increasing number of passwords for uses including host accounts, email servers, e-commerce sites, and online financial services. Unfortunately, the password entropy that users can comfortably memorize seems insufficient to store unique, secure passwords for all these accounts, and it is likely to remain constant as the number of passwords (and the adversary’s computational power) increases into the future. In this paper, we propose a technique that uses a strengthened cryptographic hash function to compute secure passwords for arbitrarily many accounts while requiring the user to memorize only a single short password. This mechanism functions entirely on the client; no server-side changes are needed. Unlike previous approaches, our design is both highly resistant to brute force attacks and nearly stateless, allowing users to retrieve their passwords from any location so long as they can execute our program and remember a short secret. This combination of security and convenience will, we believe, entice users to adopt our scheme. We discuss the construction of our algorithm in detail, compare its strengths and weaknesses to those of related approaches, and present Password Multiplier, an implementation in the form of an extension to the Mozilla Firefox web browser.

The growing popularity of inexpensive, portable recording devices, such as cellular phone cameras and compact digital audio recorders, presents a significant new threat to privacy. We propose a set of technologies that can be integrated into recording devices to provide stronger, more accurately targeted privacy protections than other legal and technical measures now under consideration. Our design is based on an informed consent principle, which it supports by the use of novel devices and protocols that automate negotiations over consent and ensure appropriate safeguards on recorded data. We define the protocols needed for this purpose and establish their security. We also describe a working prototype implementation that safeguards audio recorded by laptop PCs in a wireless network.

We explore new techniques for the use of cryptographic puzzles as a countermeasure to Denial-of-Service (DoS) attacks. We propose simple new techniques that permit the outsourcing of puzzles--their distribution via a robust external service that we call a bastion. Many servers can rely on puzzles distributed by a single bastion. We show how a bastion, somewhat surprisingly, need not know which servers rely on its services. Indeed, in one of our constructions, a bastion may consist merely of a publicly accessible random data source, rather than a special purpose server. Our outsourcing techniques help eliminate puzzle distribution as a point of compromise. Our design has three main advantages over prior approaches. First, it is more resistant to DoS attacks aimed at the puzzle mechanism itself, withstanding over 80% more attack traffic than previous methods in our experiments. Second, our scheme is cheap enough to apply at the IP level, though it also works at higher levels of the protocol stack. Third, our method allows clients to solve puzzles offline, reducing the need for users to wait while their computers solve puzzles. We present a prototype implementation of our approach, and we describe experiments that validate our performance claims.

New acquisition and modeling tools make it easier to create 3D models, and fordable and powerful graphics hardware makes it easier to use them. As a result, the number of 3D models available on the web is increasing rapidly. However, it is still not as easy to find 3D models as it is to find, for example, text documents and images. What is needed is a “3D model search engine,” a specialized search engine that targets 3D models. We created a prototype 3D model search engine to investigate the design and implementation issues. Our search engine can be partitioned into three main components: (1) acquisition: 3D models have to be collected from the web, (2) analysis: they have to be analyzed for later matching, and (3) query processing and matching: an online system has to match user queries to the collected 3D models. Our site currently indexes over 36,000 models, of which about 31,000 are freely available. In addition to a text search interface, it offers several 3D and 2D shape-based query interfaces. Since it went online one year ago (in November 2001), it has processed over 148,000 searches from 37,800 hosts in 103 different countries. Currently 20-25% of the about 1,000 visitors per week are returning users. This paper reports on our initial experiences designing, building, and running the 3D model search engine.

As the number of 3D models available on the Web grows, there is an increasing need for a search engine to help people find them. Unfortunately, traditional text-based search techniques are not always effective for 3D data. In this paper, we investigate new shape-based search methods. The key challenges are to develop query methods simple enough for novice users and matching algorithms robust enough to work for arbitrary polygonal models. We present a web-based search engine system that supports queries based on 3D sketches, 2D sketches, 3D models, and/or text keywords. For the shape-based queries, we have developed a new matching algorithm that uses spherical harmonics to compute discriminating similarity measures without requiring repair of model degeneracies or alignment of orientations. It provides 46.245% better performance than related shape matching methods during precision-recall experiments, and it is fast enough to return query results from a repository of 20,000 models in under a second. The net result is a growing interactive index of 3D models available on the Web (i.e., a Google for 3D models).

Several major record labels are adopting a new family of copy-prevention techniques intended to limit “casual” copying by compact disc owners using their personal computers. These employ deliberate data errors introduced into discs during manufacturing to cause incompatibility with PCs without affecting ordinary CD players. We examine three such recordings: A Tribute to Jim Reeves by Charley Pride, A New Day Has Come by Celine Dion, and More Music from The Fast and the Furious by various artists. In tests with different CD-ROM drives, operating systems, and playback software, we find these discs are unreadable in several widely-used applications as of July 2002. We analyze the specific technical differences between the modified recordings and standard audio CDs, and we consider repairs to hardware and software that would restore compatibility. We conclude that these schemes are harmful to legitimate CD owners and will not reduce illegal copying in the long term, so the music industry should reconsider their deployment.

For more than a century, election officials across the United States have inspected voting machines before elections using a procedure called Logic and Accuracy Testing (LAT). This procedure consists of election officials casting a test deck of ballots into each voting machine and confirming the machine produces the expected vote total for each candidate. We bring a scientific perspective to LAT by introducing the first formal approach to designing test decks with rigorous security guarantees. Specifically, our approach employs robust optimization to find test decks that are guaranteed to detect any voting machine misconfiguration that would cause votes to be swapped across candidates. Out of all the test decks with this security guarantee, our robust optimization problem yields the test deck with the minimum number of ballots, thereby minimizing implementation costs for election officials. To facilitate deployment at scale, we develop a practically efficient exact algorithm for solving our robust optimization problems based on the cutting plane method. In partnership with the Michigan Bureau of Elections, we retrospectively applied our approach to all 6928 ballot styles from Michigan's November 2022 general election; this retrospective study reveals that the test decks with rigorous security guarantees obtained by our approach require, on average, only 1.2% more ballots than current practice. Our approach has since been piloted in real-world elections by the Michigan Bureau of Elections as a low-cost way to improve election security and increase public trust in democratic institutions.

This advisory identifies vulnerabilities affecting versions of the Dominion Voting Systems Democracy Suite ImageCast X, which is an in-person voting system used to allow voters to mark their ballot. The ImageCast X can be configured to allow a voter to produce a paper record or to record votes electronically. While these vulnerabilities present risks that should be mitigated as soon as possible, CISA has no evidence that these vulnerabilities have been exploited in any elections.

Exploitation of these vulnerabilities would require physical access to individual ImageCast X devices, access to the Election Management System (EMS), or the ability to modify files before they are uploaded to ImageCast X devices. Jurisdictions can prevent and/or detect the exploitation of these vulnerabilities by diligently applying the mitigations recommended in this advisory, including technical, physical, and operational controls that limit unauthorized access or manipulation of voting systems. Many of these mitigations are already typically standard practice in jurisdictions where these devices are in use and can be enhanced to further guard against exploitation of these vulnerabilities.

Stealthy pixel-perfect attacks on smartphone apps are a class of phishing attacks that rely on visual deception to trick users into entering sensitive information into trojan apps. We introduce an operating system abstraction called Trusted Visual I/O Paths (TIVOs) that enables a user to securely verify the app she is interacting with, only assuming that the operating system provides a trusted computing base. As proof of concept, we built a TIVO for Android, one that is activated any time a soft keyboard is used by an application (e.g., for password entry) so that the user can reliably determine the app that receives the user’s keyboard input. We implemented TIVO by modifying Android’s user-interface stack and evaluated the abstraction using a controlled user study where users had to decide whether to trust the login screen of four different applications that were randomly subjected to two forms of pixel-perfect attacks. The TIVO mechanism was found to significantly reduce the effectiveness of pixel-perfect attacks, with acceptable impact on overall usability and only modest performance overhead.

Of all of the revelations about the NSA that have come to light in recent months, two stand out as the most worrisome and surprising to cybersecurity experts. The first is that the NSA has worked to weaken the international cryptographic standards that define how computers secure communications and data. The second is that the NSA has deliberately introduced backdoors into security-critical software and hardware. If the NSA has indeed engaged in such activities, it has risked the computer security of the United States (and the world) as much as any malicious attacks have to date.

Many common software vulnerabilities are avoidable if software makers apply appropriate care, yet developers’ incentives often lead them to underinvest in security. Profit-maximizing developers invest to the extent that strengthening security increases sales or reduces their liability, yet these incentives are undermined by the software market’s structure. By understanding and reshaping such incentives, we can greatly improve security at comparably low cost. The author argues for requiring increased transparency about security problems and development practices, which will help software buyers make better-informed purchases, and for holding developers liable for the costs of security failures caused by their products.

We have discovered remotely-exploitable vulnerabilities in Green Dam, the censorship software reportedly mandated by the Chinese government. Any web site a Green Dam user visits can take control of the PC. According to press reports, China will soon require all PCs sold in the country to include Green Dam. This software monitors web sites visited and other activity on the computer and blocks adult content as well as politically sensitive material. We examined the Green Dam software and found that it contains serious security vulnerabilities due to programming errors. Once Green Dam is installed, any web site the user visits can exploit these problems to take control of the computer. This could allow malicious sites to steal private data, send spam, or enlist the computer in a botnet. In addition, we found vulnerabilities in the way Green Dam processes blacklist updates that could allow the software makers or others to install malicious code during the update process. We found these problems with less than 12 hours of testing, and we believe they may be only the tip of the iceberg. Green Dam makes frequent use of unsafe and outdated programming practices that likely introduce numerous other vulnerabilities. Correcting these problems will require extensive changes to the software and careful retesting. In the meantime, we recommend that users protect themselves by uninstalling Green Dam immediately.

This report describes the hardware design of the AVC Advantage direct-recording electronic (DRE) voting machine. We developed these functional specifications by reverse engineering a government-surplus system.

MediaMax CD3 is a new copy-prevention technique from SunnComm Technologies that is designed to prevent unauthorized copying of audio CDs using personal computers. SunnComm claims its product facilitates “a verifiable and commendable level of security,” but in tests on a newly-released album, I find that the protections may have no effect on a large fraction of deployed PCs, and that most users who would be affected can bypass the system entirely by holding the shift key every time they insert the CD. I explain that MediaMax interferes with audio copying by installing a device driver the first time software from the CD is executed, but I show that this provides only minimal protection because the driver can easily be disabled. I also examine the digital rights management system used to control access to a set of encrypted, compressed audio files distributed on the CD. Although restrictions on these files are more relaxed than in prior copy protected discs, they still prohibit many uses permitted by the law. I conclude that MediaMax and similar copy-prevention systems are irreparably flawed but predict that record companies will find success with more customer-friendly alternatives for reducing infringement.