Most contemporary mobile devices offer hardware-backed storage for cryptographic keys, user data, and other sensitive credentials. Such hardware is capable of protecting credentials from extraction by an adversary who has compromised the main operating system, such as a malicious third-party app. Since 2011, Android app developers can access trusted hardware via the Android Keystore API, making hardware-backed key storage a widely accessible PET. In this work, we conduct the first comprehensive survey of hardware-backed key storage in Android devices. We analyze 490,119 Android apps, collecting data on how trusted hardware is used by app developers (if at all) and cross-referencing our findings with sensitive user data collected by each app, as self-reported by developers via the Play Store’s data safety labels. We find that despite industry-wide initiatives to encourage adoption, 56% of apps self-reporting as processing sensitive user data do not use Android’s trusted hardware capabilities at all, while just 5% of apps collecting some form of sensitive data use the strongest form of trusted hardware, a secure element distinct from the main processor. To better understand the potential downsides of using secure hardware, we conduct the first empirical analysis of trusted hardware performance in mobile devices, measuring the runtime of common cryptographic operations across both software- and hardware-backed keystores and providing the first empirical analysis to date of the performance of secure elements in mobile devices. We find that while hardware-backed key storage using a coprocessor is viable for most common cryptographic operations, secure elements capable of preventing more advanced attacks make performance infeasible for symmetric encryption with non-negligible payloads.

The advent of end-to-end encryption (E2EE) has brought new challenges for usable authentication and recovery. Compared to regular web services, the nature of E2EE requires that the provider cannot recover data for users who have forgotten passwords or lost devices. More robust recovery schemes are therefore required, leading to a plethora of solutions ranging from randomly-generated recovery codes to social authentication. These implications have spread to new forms of authentication and legacy web services: passwordless authentication (``passkeys'') has become a promising candidate to replace passwords altogether, but is inherently device-bound. However, users expect that they can login from multiple devices and recover their passwords in case of device loss---prompting providers to sync credentials to cloud storage using E2EE and making contemporary authentication for even non-E2EE services dependent on E2EE. Hence, E2EE authentication quickly becomes relevant not only for a niche group of dedicated E2EE enthusiasts but for the general public using the passwordless authentication techniques promoted by their device vendors. In this paper we systematize existing research literature and industry practice relating to security, privacy, usability, and recoverability of both end-user authentication to E2EE services and the use of E2EE in securing backend credential databases. We investigate authentication and recovery schemes in all widely-used E2EE web services, analyze syncing protocols for E2EE credential managers, and survey passwordless authentication deployment in the top-300 most popular websites. Finally, we present concrete research directions based on observed gaps between industry deployment and academic literature.

Legislators and policymakers worldwide are debating options for suppressing illegal, harmful and undesirable material online. Drawing on several quantitative data sources, we show that deplatforming an active community to suppress online hate and harassment, even with a substantial concerted effort involving several tech firms, can be hard. Our case study is the disruption of the largest and longest-running harassment forum Kiwi Farms in late 2022, which is probably the most extensive industry effort to date. Despite the active participation of a number of tech companies over several consecutive months, this campaign failed to shut down the forum and remove its objectionable content. While briefly raising public awareness, it led to rapid platform displacement and traffic fragmentation. Part of the activity decamped to Telegram, while traffic shifted from the primary domain to previously abandoned alternatives. The forum experienced intermittent outages for several weeks, after which the community leading the campaign lost interest, traffic was directed back to the main domain, users quickly returned, and the forum was back online and became even more connected. The forum members themselves stopped discussing the incident shortly thereafter, and the net effect was that forum activity, active users, threads, posts and traffic were all cut by about half. The disruption largely affected casual users (of whom roughly 87% left), while half the core members remained engaged. It also drew many newcomers, who exhibited increasing levels of toxicity during the first few weeks of participation. Deplatforming a community without a court order raises philosophical issues about censorship versus free speech; ethical and legal issues about the role of industry in online content moderation; and practical issues on the efficacy of private-sector versus government action. Deplatforming a dispersed community using a series of court orders against individual service providers appears unlikely to be very effective if the censor cannot incapacitate the key maintainers, whether by arresting them, enjoining them or otherwise deterring them.

There has been substantial commentary on the role of cyberattacks carried out by low-level cybercrime actors in the Russia-Ukraine conflict. We analyse 358k website defacement attacks, 1.7M UDP amplification DDoS attacks, 1764 posts made by 372 users on Hack Forums mentioning the two countries, and 441 Telegram announcements (with 58k replies) of a volunteer hacking group for two months before and four months after the invasion. We find the conflict briefly but notably caught the attention of low-level cybercrime actors, with significant increases in online discussion and both types of attacks targeting Russia and Ukraine. However, there was little evidence of high-profile actions; the role of these players in the ongoing hybrid warfare is minor, and they should be separated from persistent and motivated 'hacktivists' in state-sponsored operations. Their involvement in the conflict appears to have been short-lived and fleeting, with a clear loss of interest in discussing the situation and carrying out both website defacement and DDoS attacks against either Russia or Ukraine after just a few weeks.

Abstract Threat modeling is one of the foundations of secure systems engineering and must take heed of the context within which systems operate. In this work, we explore the extent to which real‐world systems engineering reflects a changing threat context. We examine the desktop clients of six widely used end‐to‐end‐encrypted mobile messaging applications to understand the extent to which they adjusted their threat model over space (when enabling clients on new platforms, such as desktop clients) and time (as new threats emerged). We experimented with short‐lived adversarial access against these desktop clients and analyzed the results using two popular threat elicitation frameworks, STRIDE and LINDDUN. The results demonstrate that system designers need to track threats in the evolving context within which systems operate and, more importantly, mitigate them by rescoping trust boundaries so that they remain consistent with administrative boundaries. A nuanced understanding of the relationship between trust and administration is vital for robust security, including the provision of safe defaults.

Search engines are vulnerable to attacks against indexing and searching via text encoding manipulation. By imperceptibly perturbing text using uncommon encoded representations, adversaries can control results across search engines for specific search queries. We demonstrate that this attack is successful against two major commercial search engines - Google and Bing - and one open source search engine - Elasticsearch. We further demonstrate that this attack is successful against LLM chat search including Bing’s GPT-4 chatbot and Google’s Bard chatbot. We also present a variant of the attack targeting text summarization and plagiarism detection models, two ML tasks closely tied to search. We provide a set of defenses against these techniques and warn that adversaries can leverage these attacks to launch disinformation campaigns against unsuspecting users, motivating the need for search engine maintainers to patch deployed systems.

Several years of research have shown that machine-learning systems are vulnerable to adversarial examples, both in theory and in practice. Until now, such attacks have primarily targeted visual models, exploiting the gap between human and machine perception. Although text-based models have also been attacked with adversarial examples, such attacks struggled to preserve semantic meaning and indistinguishability. In this paper, we explore a large class of adversarial examples that can be used to attack text-based models in a black-box setting without making any human-perceptible visual modification to inputs. We use encoding-specific perturbations that are imperceptible to the human eye to manipulate the outputs of a wide range of Natural Language Processing (NLP) systems from neural machine-translation pipelines to web search engines. We find that with a single imperceptible encoding injection – representing one invisible character, homoglyph, reordering, or deletion – an attacker can significantly reduce the performance of vulnerable models, and with three injections most models can be functionally broken. Our attacks work against currently-deployed commercial systems, including those produced by Microsoft and Google, in addition to open source models published by Facebook, IBM, and HuggingFace. This novel series of attacks presents a significant threat to many language processing systems: an attacker can affect systems in a targeted manner without any assumptions about the underlying model. We conclude that text-based NLP systems require careful input sanitization, just like conventional applications, and that given such systems are now being deployed rapidly at scale, the urgent attention of architects and operators is required.

Abstract Whistleblowing is hazardous in a world of pervasive surveillance, yet many leading newspapers expect sources to contact them with methods that are either insecure or barely usable. In an attempt to do better, we conducted two workshops with British news organisations and surveyed whistleblowing options and guidelines at major media outlets. We concluded that the soft spot is a system for initial contact and trust establishment between sources and reporters. CoverDrop is a two-way, secure system to do this. We support secure messaging within a news app, so that all its other users provide cover traffic, which we channel through a threshold mix instantiated in a Trusted Execution Environment within the news organisation. CoverDrop is designed to resist a powerful global adversary with the ability to issue warrants against infrastructure providers, yet it can easily be integrated into existing infrastructure. We present the results from our workshops, describe CoverDrop’s design and demonstrate its security and performance.

The high energy costs of neural network training and inference led to the use of acceleration hardware such as GPUs and TPUs. While such devices enable us to train large-scale neural networks in datacenters and deploy them on edge devices, their designers' focus so far is on average-case performance. In this work, we introduce a novel threat vector against neural networks whose energy consumption or decision latency are critical. We show how adversaries can exploit carefully-crafted sponge examples, which are inputs designed to maximise energy consumption and latency, to drive machine learning (ML) systems towards their worst-case performance. Sponge examples are, to our knowledge, the first denial-of-service attack against the ML components of such systems. We mount two variants of our sponge attack on a wide range of state-of-the-art neural network models, and find that language models are surprisingly vulnerable. Sponge examples frequently increase both latency and energy consumption of these models by a factor of 30×. Extensive experiments show that our new attack is effective across different hardware platforms (CPU, GPU and an ASIC simulator) on a wide range of different language tasks. On vision tasks, we show that sponge examples can be produced and a latency degradation observed, but the effect is less pronounced. To demonstrate the effectiveness of sponge examples in the real world, we mount an attack against Microsoft Azure's translator and show an increase of response time from 1ms to 6s (6000×). We conclude by proposing a defense strategy: shifting the analysis of energy consumption in hardware from an average-case to a worst-case perspective.

Inpainting is a learned interpolation technique that is based on generative modeling and used to populate masked or missing pieces in an image; it has wide applications in picture editing and retouching. Recently, inpainting started being used for watermark removal, raising concerns. In this paper we study how to manipulate it using our markpainting technique. First, we show how an image owner with access to an inpainting model can augment their image in such a way that any attempt to edit it using that model will add arbitrary visible information. We find that we can target multiple different models simultaneously with our technique. This can be designed to reconstitute a watermark if the editor had been trying to remove it. Second, we show that our markpainting technique is transferable to models that have different architectures or were trained on different datasets, so watermarks created using it are difficult for adversaries to remove. Markpainting is novel and can be used as a manipulation alarm that becomes visible in the event of inpainting. Source code is available at: https://github.com/iliaishacked/markpainting.

Machine learning is vulnerable to a wide variety of attacks. It is now well understood that by changing the underlying data distribution, an adversary can poison the model trained with it or introduce backdoors. In this paper we present a novel class of training-time attacks that require no changes to the underlying dataset or model architecture, but instead only change the order in which data are supplied to the model. In particular, we find that the attacker can either prevent the model from learning, or poison it to learn behaviours specified by the attacker. Furthermore, we find that even a single adversarially-ordered epoch can be enough to slow down model learning, or even to reset all of the learning progress. Indeed, the attacks presented here are not specific to the model or dataset, but rather target the stochastic nature of modern learning procedures. We extensively evaluate our attacks on computer vision and natural language benchmarks to find that the adversary can disrupt model training and even introduce backdoors.

Security engineers have been fighting with C compilers for years. A careful programmer would test for null pointer dereferencing or division by zero; but the compiler would fail to understand, and optimize the test away. Modern compilers now have dedicated options to mitigate this. But when a programmer tries to control side effects of code, such as to make a cryptographic algorithm execute in constant time, the problem remains. Programmers devise complex tricks to obscure their intentions, but compiler writers find ever smarter ways to optimize code. A compiler upgrade can suddenly and without warning open a timing channel in previously secure code. This arms race is pointless and has to stop. We argue that we must stop fighting the compiler, and instead make it our ally. As a starting point, we analyze the ways in which compiler optimization breaks implicit properties of crypto code; and add guarantees for two of these properties in Clang/LLVM. Our work explores what is actually involved in controlling side effects on modern CPUs with a standard toolchain. Similar techniques can and should be applied to other security properties; achieving intentions by compiler commands or annotations makes them explicit, so we can reason about them. It is already understood that explicitness is essential for cryptographic protocol security and for compiler performance; it is essential for language security too. We therefore argue that this should be only the first step in a sustained engineering effort.

At the first Information Hiding Workshop in 1996 we tried to clarify the models and assumptions behind information hiding. We agreed the terminology of cover text and stego text against a background of the game proposed by our keynote speaker Gus Simmons: that Alice and Bob are in jail and wish to hatch an escape plan without the fact of their communication coming to the attention of the warden, Willie. Since then there have been significant strides in developing technical mechanisms for steganography and steganalysis, with new techniques from machine learning providing ever more powerful tools for the analyst, such as the ensemble classifier. There have also been a number of conceptual advances, such as the square root law and effective key length. But there always remains the question whether we are using the right security metrics for the application. In this talk I plan to take a step backwards and look at the systems context. When can stegosystems actually be used? The deployment history is patchy, with one being Trucrypt's hidden volumes, inspired by the steganographic file system. Image forensics also find some use, and may be helpful against some adversarial machine learning attacks (or at least help us understand them). But there are other contexts in which patterns of activity have to be hidden for that activity to be effective. I will discuss a number of examples starting with deception mechanisms such as honeypots, Tor bridges and pluggable transports, which merely have to evade detection for a while; then moving on to the more challenging task of designing deniability mechanisms, from leaking secrets to a newspaper through bitcoin mixes, which have to withstand forensic examination once the participants come under suspicion. We already know that, at the system level, anonymity is hard. However the increasing quantity and richness of the data available to opponents may move a number of applications from the deception category to that of deniability. To pick up on our model of 20 years ago, Willie might not just put Alice and Bob in solitary confinement if he finds them communicating, but torture them or even execute them. Changing threat models are historically one of the great disruptive forces in security engineering. This leads me to suspect that a useful research area may be the intersection of deception and forensics, and how information hiding systems can be designed in anticipation of richer and more complex threat models. The ever-more-aggressive censorship systems deployed in some parts of the world also raise the possibility of using information hiding techniques in censorship circumvention. As an example of recent practical work, I will discuss Covertmark, a toolkit for testing pluggable transports that was partly inspired by Stirmark, a tool we presented at the second Information Hiding Workshop twenty years ago.

Summary form only given. As people get excited about the latest idea for "Big Data" and the "Internet of Things", computer people often shake our heads and say "It won't scale." Pessimism isn't always justified: we have been able to scale up quite a number of tasks, from connectivity through search to social media. But other applications are recalcitrant, from energy management to medical records. The conventional computer-science view is that scaling systems is about computational complexity; about whether the storage or communications required for a task grows more than linearly in the number of users. Over the past thirty years we've developed a pretty good theory of that, but we're learning that it's nowhere near enough. In this talk I present a complementary view, based on over thirty years' experience of security engineering, that the real limits to scale are usually elsewhere. Even where the data are manageable and the algorithms straightforward, things can fail because of the scaling properties of the social context, the economic model or the regulatory environment. This makes some automation projects much harder than they seem. When it comes to safety and privacy many of the attacks that are easy to do in the lab are rare in the wild, as they don't scale either. But others surprise us; no-one in the intelligence community anticipated a leak on the Snowden scale. In short, scaling is now a problem not of computer science but of systems engineering, economics, governance and much else. Conceiving problems too narrowly makes failure likely, while good engineering will require ever more awareness of context. The implications for research, education and policy bear some thought.

Abstract We present a new side-channel attack against soft keyboards that support gesture typing on Android smartphones. An application without any special permissions can observe the number and timing of the screen hardware interrupts and system-wide software interrupts generated during user input, and analyze this information to make inferences about the text being entered by the user. System-wide information is usually considered less sensitive than app-specific information, but we provide concrete evidence that this may be mistaken. Our attack applies to all Android versions, including Android M where the SELinux policy is tightened. We present a novel application of a recurrent neural network as our classifier to infer text. We evaluate our attack against the “Google Keyboard” on Nexus 5 phones and use a real-world chat corpus in all our experiments. Our evaluation considers two scenarios. First, we demonstrate that we can correctly detect a set of pre-defined “sentences of interest” (with at least 6 words) with 70% recall and 60% precision. Second, we identify the authors of a set of anonymous messages posted on a messaging board. We find that even if the messages contain the same number of words, we correctly re-identify the author more than 97% of the time for a set of up to 35 sentences. Our study demonstrates a new way in which system-wide resources can be a threat to user privacy. We investigate the effect of rate limiting as a countermeasure but find that determining a proper rate is error-prone and fails in subtle cases. We conclude that real-time interrupt information should be made inaccessible, perhaps via a tighter SELinux policy in the next Android version.

EMV, also known as "Chip and PIN", is the leading system for card payments worldwide. It is used throughout Europe and much of Asia, and is starting to be introduced in North America too. Payment cards contain a chip so they can execute an authentication protocol. This protocol requires point-of-sale (POS) terminals or ATMs to generate a nonce, called the unpredictable number, for each transaction to ensure it is fresh. We have discovered two serious problems: a widespread implementation flaw and a deeper, more difficult to fix flaw with the EMV protocol itself. The first flaw is that some EMV implementers have merely used counters, timestamps or home-grown algorithms to supply this nonce. This exposes them to a "pre-play" attack which is indistinguishable from card cloning from the standpoint of the logs available to the card-issuing bank, and can be carried out even if it is impossible to clone a card physically. Card cloning is the very type of fraud that EMV was supposed to prevent. We describe how we detected the vulnerability, a survey methodology we developed to chart the scope of the weakness, evidence from ATM and terminal experiments in the field, and our implementation of proof-of-concept attacks. We found flaws in widely-used ATMs from the largest manufacturers. We can now explain at least some of the increasing number of frauds in which victims are refused refunds by banks which claim that EMV cards cannot be cloned and that a customer involved in a dispute must therefore be mistaken or complicit. The second problem was exposed by the above work. Independent of the random number quality, there is a protocol failure: the actual random number generated by the terminal can simply be replaced by one the attacker used earlier when capturing an authentication code from the card. This variant of the pre-play attack may be carried out by malware in an ATM or POS terminal, or by a man-in-the-middle between the terminal and the acquirer. We explore the design and implementation mistakes that enabled these flaws to evade detection until now: shortcomings of the EMV specification, of the EMV kernel certification process, of implementation testing, formal analysis, and monitoring customer complaints. Finally we discuss countermeasures. More than a year after our initial responsible disclosure of these flaws to the banks, action has only been taken to mitigate the first of them, while we have seen a likely case of the second in the wild, and the spread of ATM and POS malware is making it ever more of a threat.

The problem of matching between binaries is important for software copyright enforcement as well as for identifying disclosed vulnerabilities in software. We present a search engine prototype called Rendezvous which enables indexing and searching for code in binary form. Rendezvous identifies binary code using a statistical model comprising instruction mnemonics, control flow sub-graphs and data constants which are simple to extract from a disassembly, yet normalising with respect to different compilers and optimisations. Experiments show that Rendezvous achieves F2measures of 86.7% and 83.0% on the GNU C library compiled with different compiler optimisations and the GNU coreutils suite compiled with gcc and clang respectively. These two code bases together comprise more than one million lines of code. Rendezvous will bring significant changes to the way patch management and copyright enforcement is currently performed.

This paper describes the origins of security economics. The birth of this thriving new discipline is sometimes credited to a talk I gave at ACSAC in December 2001, but the story is more complex. After sabbatical visits to Berkeley in 2001--2 to work with Hal Varian, we organised the first Workshop on the Economics of Information Security in June 2002. Since then the field has grown to encompass arguments over open versus proprietary systems, the econometrics of online crime, the behavioural economics of security and much else. It has started to have a significant impact on policy, with security-economics studies of cybercrime and infrastructure vulnerability being adopted as policy in the EU, while security economics PhDs have got influential jobs in the White House and elsewhere.

EMV is the dominant protocol used for smart card payments worldwide, with over 730 million cards in circulation. Known to bank customers as “Chip and PIN”, it is used in Europe; it is being introduced in Canada; and there is pressure from banks to introduce it in the USA too. EMV secures credit and debit card transactions by authenticating both the card and the customer presenting it through a combination of cryptographic authentication codes, digital signatures, and the entry of a PIN. In this paper we describe and demonstrate a protocol flaw which allows criminals to use a genuine card to make a payment without knowing the card's PIN, and to remain undetected even when the merchant has an online connection to the banking network. The fraudster performs a man-in-the-middle attack to trick the terminal into believing the PIN verified correctly, while telling the card that no PIN was entered at all. The paper considers how the flaws arose, why they remained unknown despite EMV's wide deployment for the best part of a decade, and how they might be fixed. Because we have found and validated a practical attack against the core functionality of EMV, we conclude that the protocol is broken. This failure is significant in the field of protocol design, and also has important public policy implications, in light of growing reports of fraud on stolen EMV cards. Frequently, banks deny such fraud victims a refund, asserting that a card cannot be used without the correct PIN, and concluding that the customer must be grossly negligent or lying. Our attack can explain a number of these cases, and exposes the need for further research to bridge the gap between the theoretical and practical security of bank payment systems. It also demonstrates the need for the next version of EMV to be engineered properly.

There has been some discussion in the industrial control system security community of evaluation and certification. There are already at least two independent third party evaluators, and some have advocated common criteria certification of products used in critical systems. The broader IT security community has considerable experience of evaluation and certification, which we seek to summarise and share in this paper. Certification is not a silver bullet, and can very easily end up as spin rather than substance: as `security theatre' designed to reassure customers or regulators rather than a genuine risk-reduction mechanism. It can also be very expensive, and once entrenched it can impose deadweight costs on industry that are difficult to eliminate even when certification processes are widely seen as failing. We discuss a number of further issues such as perverse incentives, usability and liability and argue that the industry should proceed with great caution.

Vehicular networks, in which cars communicate wirelessly to exchange information on traffic conditions, offer a promising way to improve road safety. Yet ensuring the correct functioning of such a system is essential: malicious or faulty devices transmitting inaccurate messages could trigger accidents. Therefore, any errant device, along with the messages it generates, must be identified and ignored as quickly as possible. This task is especially challenging because traditional approaches to revoking credentials use a central authority, causing long delays during which the network is vulnerable. To eliminate this window of vulnerability, we propose that vehicles locally decide whether to exclude errant devices. We describe two ways of doing so: first, LEAVE, an existing protocol which allows devices to vote by exchanging signed claims of impropriety, and second, Stinger, a new protocol where a device unilaterally removes a misbehaving neighbor by agreeing to limit its own participation. We provide detailed simulations that offer insight into the protocols' operations in the context of vehicular networks and enable a powerful comparison between the strategies. We compare the security and performance properties of LEAVE and Stinger while varying attacker capabilities, traffic conditions, and the accuracy of the misbehavior detection mechanisms. We identify several interesting trade-offs: Stinger is significantly faster than LEAVE at removing errant devices, but LEAVE excludes fewer good devices when the attacker has compromised several devices simultaneously; LEAVE is better at handling false positives, but Stinger scales better when the traffic density increases. As a result, we conclude by outlining a combined protocol that balances the security and performance characteristics of both strategies.

PIN entry devices (PEDs) are critical security components in EMV smartcard payment systems as they receive a customer's card and PIN. Their approval is subject to an extensive suite of evaluation and certification procedures. In this paper, we demonstrate that the tamper proofing of PEDs is unsatisfactory, as is the certification process. We have implemented practical low-cost attacks on two certified, widely-deployed PEDs - the Ingenico 13300 and the Dione Xtreme. By tapping inadequately protected smartcard communications, an attacker with basic technical skills can expose card details and PINs, leaving cardholders open to fraud. We analyze the anti-tampering mechanisms of the two PEDs and show that, while the specific protection measures mostly work as intended, critical vulnerabilities arise because of the poor integration of cryptographic, physical and procedural protection. As these vulnerabilities illustrate a systematic failure in the design process, we propose a methodology for doing it better in the future. These failures also demonstrate a serious problem with the Common Criteria. So we discuss the incentive structures of the certification process, and show how they can lead to problems of the kind we identified. Finally we recommend changes to the Common Criteria framework in light of the lessons learned.

In this paper we describe the protection goals and mechanisms in HomePlug AV, a next-generation power-line communications standard. This is a fascinating case-history in security usability. There are also novel protocol issues; interactions with mechanisms at other layers; and opportunities for both researchers and third-party vendors to build on the mechanisms provided. The central problem -- being sure whether a device being enrolled in the network is the device you think, not a similar one nearby -- is not well solved by conventional mechanisms such as public-key infrastructures, but appears to require either very old-fashioned or very novel approaches.

Tamper-resistant cryptographic processors are becoming the standard way to enforce data-usage policies. Their origins lie with military cipher machines and PIN processing in banking payment networks, expanding in the 1990s into embedded applications: token vending machines for prepayment electricity and mobile phone credit. Major applications such as GSM mobile phone identification and pay TV set-top boxes have pushed low-cost cryptoprocessors toward ubiquity. In the last five years, dedicated crypto chips have been embedded in devices such as game console accessories and printer ink cartridges, to control product and accessory after markets. The "Trusted Computing" initiative will soon embed cryptoprocessors in PCs so they can identify each other remotely. This paper surveys the range of applications of tamper-resistant hardware and the array of attack and defense mechanisms which have evolved in the tamper-resistance arms race.

We propose the first practical and secure way to integrate the iris biometric into cryptographic applications. A repeatable binary string, which we call a biometric key, is generated reliably from genuine iris codes. A well-known difficulty has been how to cope with the 10 to 20 percent of error bits within an iris code and derive an error-free key. To solve this problem, we carefully studied the error patterns within iris codes and devised a two-layer error correction technique that combines Hadamard and Reed-Solomon codes. The key is generated from a subject's iris image with the aid of auxiliary error-correction data, which do not reveal the key and can be saved in a tamper-resistant token, such as a smart card. The reproduction of the key depends on two factors: the iris biometric and the token. The attacker has to procure both of them to compromise the key. We evaluated our technique using iris samples from 70 different eyes, with 10 samples from each eye. We found that an error-free key can be reproduced reliably from genuine iris codes with a 99.5 percent success rate. We can generate up to 140 bits of biometric key, more than enough for 128-bit AES. The extraction of a repeatable binary string from biometrics opens new possible applications, where a strong binding is required between a person and cryptographic operations. For example, it is possible to identify individuals without maintaining a central database of biometric templates, to which privacy objections might be raised.

Future distributed systems may include large self-organizing networks of locally communicating sensor nodes, any small number of which may be subvened by an adversary. Providing security for these sensor networks is important, but the problem is complicated by the fact that managing cryptographic key material is hard: low-cost nodes are neither tamper-proof nor capable of performing public key cryptography efficiently. We show how the key distribution problem can be dealt with in environments with a partially present, passive adversary: a node wishing to communicate securely with other nodes simply generates a symmetric key and sends it in the clear to its neighbours. Despite the apparent insecurity of this primitive, we can use mechanisms for key updating, multipath secrecy amplification and multihop key propagation to build up extremely resilient trust networks where at most a fixed proportion of communications links can be eavesdropped. We discuss applications in which this assumption is sensible. Many systems must perforce cope with principals who are authenticated weakly, if at all; the resulting issues have often been left in the 'too hard' tray. One particular interest of sensor networks is that they present a sufficiently compact and tractable version of this problem. We can perform quantitative analyses and simulations of alternative strategies, some of which we present here. We also hope that This work may start to challenge the common belief that authentication is substantially about bootstrapping trust. We argue that, in distributed systems where the opponent can subvert any small proportion of nodes, it is more economic to invest in resilience than in bootstrapping.

The most significant strategic development in information technology over the past year has been 'trusted computing'. This is popularly associated with Microsoft's 'Palladium' project, recently renamed 'NGSCB'. In this paper, I give an outline of the technical aspects of 'trusted computing' and sketch some of the public policy consequences.

According to one common view, information security comes down to technical measures. Given better access control policy models, formal proofs of cryptographic protocols, approved firewalls, better ways of detecting intrusions and malicious code, and better tools for system evaluation and assurance, the problems can be solved. The author puts forward a contrary view: information insecurity is at least as much due to perverse incentives. Many of the problems can be explained more clearly and convincingly using the language of microeconomics: network externalities, asymmetric information, moral hazard, adverse selection, liability dumping and the tragedy of the commons.

The probability that information will be abused depends both on its value and on the number of people, who have access. The modern trend to ever larger databases increases both of these risk factors at the same time. Compartmented security policies can solve many of the technical issues, and there are applications such as healthcare where they have been developed in some detail. But the big problem isn't technical; it is legal and regulatory. Insurers, employers and governments won't adopt compartmented systems, or will allow them to be adopted only in places such as hospitals which are not where the real threats lie.

Collaborative software projects such as Linux and Apache have shown that a large, complex system can be built and maintained by many developers working in a highly parallel, relatively unstructured way. The author reports on an experiment to see whether a high quality system specification can also be produced by a large number of people working in parallel with minimum communication.

Information-hiding techniques have recently become important in a number of application areas. Digital audio, video, and pictures are increasingly furnished with distinguishing but imperceptible marks, which may contain a hidden copyright notice or serial number or even help to prevent unauthorized copying directly. Military communications systems make increasing use of traffic security techniques which, rather than merely concealing the content of a message using encryption, seek to conceal its sender, its receiver, or its very existence. Similar techniques are used in some mobile phone systems and schemes proposed for digital elections. Criminals try to use whatever traffic security properties are provided intentionally or otherwise in the available communications systems, and police forces try to restrict their use. However, many of the techniques proposed in this young and rapidly evolving field can trace their history back to antiquity, and many of them are surprisingly easy to circumvent. In this article, we try to give an overview of the field, of what we know, what works, what does not, and what are the interesting topics for research.

In this paper, we clarify what steganography is and what it can do. We contrast it with the related disciplines of cryptography and traffic security, present a unified terminology agreed at the first international workshop on the subject, and outline a number of approaches-many of them developed to hide encrypted copyright marks or serial numbers in digital audio or video. We then present a number of attacks, some new, on such information hiding schemes. This leads to a discussion of the formidable obstacles that lie in the way of a general theory of information hiding systems (in the sense that Shannon gave us a general theory of secrecy systems). However, theoretical considerations lead to ideas of practical value, such as the use of parity checks to amplify covertness and provide public key steganography. Finally, we show that public key information hiding systems exist, and are not necessarily constrained to the case where the warden is passive.

The protection of personal health information has become a live issue in a number of countries, including the USA, Canada, Britain and Germany. The debate has shown that there is widespread confusion about what should be protected, and why. Designers of military and banking systems can refer to Bell & LaPadula (1973) and Clark & Wilson (1987) respectively, but there is no comparable security policy model that spells out clear and concise access rules for clinical information systems. In this article, we present just such a model. It was commissioned by doctors and is driven by medical ethics; it is informed by the actual threats to privacy, and reflects current best clinical practice. Its effect is to restrict both the number of users who can access any record and the maximum number of records accessed by any user. This entails controlling information flows across rather than down and enforcing a strong notification property. We discuss its relationship with existing security policy models, and its possible use in other applications where information exposure must be localised; these range from private banking to the management of intelligence data.

One of the problems facing the builders of the 'Information Superhighway' is how to charge for services. The high costs of billing systems suggest that prepayment mechanisms could play a large part in the solution. Yet how does one go about making an electronic prepayment system (or indeed any kind of payment system) robust? We describe some recent systems engineering experience which may be relevant-the successful introduction of cryptology to protect prepayment electricity meters from token fraud. These meters are used by a number of utilities from Scotland to South Africa, and they present some interesting reliability challenges.

We describe the successful introduction of cryptology into a new application area-protecting prepayment electricity meters from token fraud. These meters are used by a number of utilities from Scotland to South Africa, and they present some interesting security challenges.>

There are three widely held beliefs about cryptology which underpin much research on the subject, guide substantial industrial investment and inform most of the associated public policy debate. These are, first, that the cryptographic systems are mostly used to keep communications secure, in the sense of ensuring that an electronic message is secret and/or authentic. In the second of these roles, cryptology provides a useful (if not only) means of securing electronic evidence in such a way that it will be accepted by a court, and is thus indispensable to the future development of electronic commerce. Third, most attacks on cryptographic systems involve technical skill at cryptanalysis, and such systems can therefore be made sufficiently robust by using standard, well studied encryption and signature algorithms. Shows that these three beliefs are almost completely mistaken.

Designers of cryptographic systems are at a disadvantage to most other engineers, in that information on how their systems fail is hard to get: their major users have traditionally been government agencies, which are very secretive about their mistakes.