Large language models (LLMs) are rapidly being adopted for tasks like drafting emails, summarizing meetings, and answering health questions.In these settings, users may need to share private information (e.g., contact details, health records).To evaluate LLMs' ability to identify and redact such information, prior work introduced real-life, scenario-based benchmarks (e.g., ConfAIde, PrivacyLens) and found that LLMs can leak private information in complex scenarios.However, these evaluations relied on proxy LLMs to judge the helpfulness and privacy-preservation quality of LLM responses, rather than directly measuring users' perceptions.To understand how users perceive the helpfulness and privacy-preservation quality of LLM responses to privacy-sensitive scenarios, we conducted a user study (n = 94) using 90 PrivacyLens scenarios.We found that users had low agreement with each other when evaluating identical LLM responses.In contrast, five proxy LLMs reached high agreement, yet each proxy LLM had low correlation with users' evaluations.These results indicate that proxy LLMs cannot accurately estimate users' wide range of perceptions of utility and privacy in privacy-sensitive scenarios.We discuss the need for more user-centered studies to measure LLMs' ability to help users while preserving privacy, and for improving alignment between LLMs and users in estimating perceived privacy and utility.

Traditional estate planning practices enable people to provide their heirs access to the assets left behind but are often insufficient for the transfer and management of online accounts. To understand how estate planning practices could be improved, we conducted 21 semi-structured interviews with older adults in the United States that explored their practices, concerns, and needs regarding postmortem online account access and management. We encountered few formalized digital estate planning practices; many participants use their credential management practices—primarily pen-and-paper—to provide postmortem account access. How participants envision account transfer is motivated by trust in their current practices and in their heirs, while concerns regarding technology hinder adoption of new methods. Participants consistently prioritize accounts with financial assets, and expectations surrounding postmortem account management vary based on individual circumstances, with the common goal of reducing burdens on executors and heirs. Our results suggest the need for developing technical standardization and expert guidance for digital estate planning.

Smart-home automations enable users to customize smart devices to react automatically to people, the environment, and more. For example, an automation might adjust the lights when people are at home or enable a garage door to open by voice command. While automations offer convenience and accessibility, they can also inadvertently expose users to security and privacy risks, such as leaking sensitive data or allowing untrusted parties to control users' devices. Prior work has shown that information flow analysis is a promising technique for identifying these kinds of risks, hypothesizing that the analysis would be yet more effective if it could differentiate between devices located in different places in the home. We tested this hypothesis by developing a tool that extends prior information flow analysis approaches to account for device location. We conducted an interview study with 22 participants to build a dataset of home automations to establish a ground truth to evaluate the tool. We found that incorporating device location leads to an improved analysis that identifies more of the vulnerabilities users care about (F1 score 0.74) compared to prior work (F1 score 0.29). Our results demonstrate the feasibility of incorporating device location into an information flow analysis and, perhaps more importantly, suggest additional ways to prevent security and privacy risks beyond controlling potentially unsafe information flows.

Large language models (LLMs) are prone to hallucinations and sensitive to prompt perturbations, often resulting in inconsistent or unreliable generated text.Different methods have been proposed to mitigate such hallucinations and fragility, one of which is to measure the consistency of LLM responses-the model's confidence in the response or likelihood of generating a similar response when resampled.In previous work, measuring LLM response consistency often relied on calculating the probability of a response appearing within a pool of resampled responses, analyzing internal states, or evaluating logits of resopnses.However, it was not clear how well these approaches approximated users' perceptions of consistency of LLM responses.To find out, we performed a user study (n = 2, 976) demonstrating that current methods for measuring LLM response consistency typically do not align well with humans' perceptions of LLM consistency.We propose a logit-based ensemble method for estimating LLM consistency and show that our method matches the performance of the bestperforming existing metric in estimating human ratings of LLM consistency.Our results suggest that methods for estimating LLM consistency without human evaluation are sufficiently imperfect to warrant broader use of evaluation with human input; this would avoid misjudging the adequacy of models because of the imperfections of automated consistency metrics.

Cyber deception aims to distract, delay, and detect network attackers with fake assets such as honeypots, decoy credentials, or decoy files. However, today, it is difficult for operators to experiment, explore, and evaluate deception approaches. Existing tools and platforms have non-portable and complex implementations that are difficult to modify and extend. We address this pain point by introducing Perry, a highlevel framework that accelerates the design and exploration of deception what-if scenarios. Perry has two components: a highlevel abstraction layer for security operators to specify attackers and deception strategies and an experimentation module to run these attackers and defenders in realistic emulated networks. To translate these high-level specifications into low-level primitives, we design four key modules in Perry: 1) an action planner that translates high-level actions into low-level implementations, 2) an observability module to translate low-level telemetry into highlevel observations, 3) an environment state service that enables environment agnostic strategies, and 4) an attack graph service to reason how attackers could explore an environment. We illustrate that Perry’s abstractions reduce the implementation effort across a wide variety of deception defenses, attackers, and environments. We demonstrate the value of Perry by emulating 55 unique deception what-if scenarios, illustrating how these experiments enable operators to shed light on subtle tradeoffs.

Machine-learning (ML) classifiers are increasingly used to distinguish malware from benign binaries. Recent work has shown that ML-based detectors can be evaded by adversarial examples, but also that one may defend against such attacks via adversarial training. However, adversarial training, and subsequent robustness evaluation, is computationally expensive in the raw-binary malware-detection domain because it requires producing many adversarial examples for both training and evaluation. Prior work found that Greedy-training, a faster robust training technique that forgoes using adversarial examples, showed some promise in producing robust malware detectors. However, Greedy-training was far less effective in inducing robustness than the more expensive adversarial training, and it also severely hurt natural accuracy (i.e., accuracy on the original data). To faster train models, this work presents GreedyBlock-training, an enhanced version of Greedy-training that we empirically show achieves not only state-of-the-art robustness in malware detectors, exceeding even adversarial training, but also retains natural accuracy better than adversarial training. Furthermore, as it does not require creating adversarial (or functional) examples, GreedyBlock-training is significantly faster than adversarial training. Specifically, we show that GreedyBlock-training can produce more robust (+54% on average), more naturally accurate (+7% on average), and more efficiently trained (-91% average computation) malware detectors than prior work. To faster evaluate models, we also develop methods to faster gauge the robustness of ML-based raw-binary malware detectors by introducing robustness proxies, which can be used either to predict which models are likely to be the most robust, thus helping prioritize which detectors to evaluate with expensive attacks, or aiding in deciding which detectors are worthwhile to continue training. Experimentally, we show these proxy measures can find the most robust detector in a pool of detectors while using only ~20-50% of the computation that would otherwise be required.

As energy infrastructure becomes more interconnected, understanding cybersecurity risks to production systems requires integrating operational and computer security knowledge. We interviewed 18 experts working in the field of energy critical infrastructure to compare what information they find necessary to assess the impact of computer vulnerabilities on energy operational technology. These experts came from two groups: 1) computer security experts and 2) energy sector operations experts. We find that both groups responded similarly for general categories of information and displayed knowledge about both domains, perhaps due to their interdisciplinary work at the same organization. Yet, we found notable differences in the details of their responses and in their stated perceptions of each group’s approaches to impact assessment. Their suggestions for collaboration across domains highlighted how these two groups can work together to help each other secure the energy grid. Our findings inform the development of interdisciplinary security approaches in critical-infrastructure contexts.

Contact tracing can limit the spread of infectious diseases by notifying people of potential exposure to disease. Manual contact tracing is resource-intensive, but much of it can be automated using mobile phones, which are ubiquitous and can detect and record nearby contacts. Two major problems arise with automated contact tracing (ACT): preventing abuse for mass surveillance and accurately determining contacts. For example, the most widely adopted solution---Google and Apple's Exposure Notification (GAEN)---protects user privacy but suffers from inaccurate distance measurements that result in poor risk assessments. We propose to use collaboration among nearby devices to increase distance estimation accuracy, and therefore risk assessment accuracy, while minimizing the loss of user privacy. Our protocol, CoCoT, extends GAEN, a proximity-based, distributed ACT protocol, by adding an additional broadcast to share locally-derived distance estimates. To evaluate CoCoT, we develop a method for merging phone sensor datasets with human interaction datasets to approximate realistic scenarios and test our protocol. CoCoT improves distance estimate accuracy by 28% over the current best distance estimators and we analytically show impact on privacy, security, and battery consumption are minimal.

In the field of aviation, the Detect and Avoid (DAA) problem deals with incorporating collision avoidance capabilities into current autopilot navigation systems. As an application of the Small Object Detection (SOD) problem, DAA presents the difficulties of a low signal-to-noise ratio and far range detection. Visual DAA is also susceptible to changing weather and lighting conditions at deployment. While current literature has presented many solutions for this, prior work has yet to study the robustness of the learning-based models for DAA. In this work, we show that standard techniques for improving robustness for object detection do not produce the desired results for DAA given the SOD constraints. We present targeted transformations, a zero-shot technique that can significantly improve robustness with minimal impact on accuracy. We demonstrate how to construct these transformations and evaluate our method on the current SOTA model for DAA, showing a 53.6% increase in recall. This makes our pipeline more robust to changes in lighting and environmental factors, and better able to detect potential threats. In the future, we hope to automate the transformation selection process, making it easier to adopt in different use cases.

Randomized smoothing is a leading approach for constructing classifiers that are certifiably robust against adversarial examples. Existing work on randomized smoothing has focused on classifiers with continuous inputs, such as images, where $\ell_p$-norm bounded adversaries are commonly studied. However, there has been limited work for classifiers with discrete or variable-size inputs, such as for source code, which require different threat models and smoothing mechanisms. In this work, we adapt randomized smoothing for discrete sequence classifiers to provide certified robustness against edit distance-bounded adversaries. Our proposed smoothing mechanism randomized deletion (RS-Del) applies random deletion edits, which are (perhaps surprisingly) sufficient to confer robustness against adversarial deletion, insertion and substitution edits. Our proof of certification deviates from the established Neyman-Pearson approach, which is intractable in our setting, and is instead organized around longest common subsequences. We present a case study on malware detection—a binary classification problem on byte sequences where classifier evasion is a well-established threat model. When applied to the popular MalConv malware detection model, our smoothing mechanism RS-Del achieves a certified accuracy of 91% at an edit distance radius of 128 bytes.

The operational, academic, and policy communities disagree on which threats against the power grid are likely and what damage would ensue. For instance, the feasibility and impact of MadIoT-style attacks is being actively debated. By surveying grid experts (N=18) we find that disagreements are not unique to MadIoT attacks but occur across multiple well-studied grid threats. Based on prior work and our survey, we hypothesize that the disagreements stem from inconsistencies in how grid threats are modeled. We identify five likely causes of modeling inconsistencies: 1) using unrealistic grid topologies, 2) assuming unrealistic capabilities for attackers, 3) exploring too few grid scenarios, 4) using incomplete simulators that omit relevant grid processes, and 5) using simulators that incorrectly model key grid processes. To check these hypotheses, we create a modeling framework and examine how these factors change our understanding of the feasibility and impact of grid threats. We use four diverse grid threats as case studies: MadIoT, False Data Injection Attacks, Substation Circuit Breaker Takeover, and Power Plant Takeover. We find that each of our hypothe-sized causes of modeling inconsistencies has a significant effect on modeling the outcomes of attacks. For example, we find that MadIoT attacks are much less feasible and require significantly more high-wattage IoT devices on realistic topologies than on topologies previously used to model them. In contrast, we find that Substation Circuit Breaker Takeover attacks are much more feasible in emergency scenarios and may require significantly fewer substations for failure than previous modeling suggested. We conclude with actionable recommendations for accurately assessing the impact of threats against the grid.

As technology companies develop mass market augmented reality (AR) glasses that are increasingly sensor-laden and affordable, uses of such devices pose potential privacy and security problems. Though prior work has broadly addressed some of these problems, our work specifically addresses the potential data collection of 15 data types by AR glasses and five potential data uses. Via semi-structured interviews, we explored the attitudes and concerns of 21 current AR technology users regarding potential data collection and data use by hypothetical consumer-grade AR glasses. Participants expressed diverse concerns and suggested potential limits to AR data collection and use, evoking privacy concepts and informational norms. We discuss how participants’ attitudes and reservations about data collection and use, like definitions of privacy, are varying and context-dependent, and make recommendations for designers and policy makers, including customizable and multidimensional privacy solutions.

We propose new, more efficient targeted white-box attacks against deep neural networks. Our attacks better align with the attacker’s goal: (1) tricking a model to assign higher probability to the target class than to any other class, while (2) staying within an $\epsilon$-distance of the attacked input. First, we demonstrate a loss function that explicitly encodes (1) and show that Auto-PGD finds more attacks with it. Second, we propose a new attack method, Constrained Gradient Descent (CGD), using a refinement of our loss function that captures both (1) and (2). CGD seeks to satisfy both attacker objectives—misclassification and bounded $\ell_{p}$-norm—in a principled manner, as part of the optimization, instead of via ad hoc post-processing techniques (e.g., projection or clipping). We show that CGD is more successful on CIFAR10 (0.9–4.2%) and ImageNet (8.6–13.6%) than state-of-the-art attacks while consuming less time (11.4–18.8%). Statistical tests confirm that our attack outperforms others against leading defenses on different datasets and values of $\epsilon$.

Ad blockers heavily rely on filter lists to block ad domains, which can serve advertisements and trackers. However, recent research has reported that some advertisers keep registering replica ad domains (RAD domains)—new domains that serve the same purpose as the original ones—which tend to slip through ad-blocker filter lists. Although this phenomenon might negatively affect ad blockers’ effectiveness, no study to date has thoroughly investigated its prevalence and the issues caused by RAD domains. In this work, we proposed methods to discover RAD domains and categorized their change patterns. From a crawl of 50,000 websites, we identified 1,748 unique RAD domains, 1,096 of which survived for an average of 410.5 days before they were blocked; the rest have not been blocked as of February 2021. Notably, we found that non-blocked RAD domains could extend the timespan of ad or tracker distribution by more than two years. Our analysis further revealed a taxonomy of four techniques used to create RAD domains, including two less-studied ones. Additionally, we discovered that the RAD domains affected 10.2% of the websites we crawled, and 23.7% of the RAD domains exhibiting privacy-intrusive behaviors, undermining ad blockers’ privacy protection.

In order to keep one's computing systems and data secure, it is critical to be aware of how to effectively maintain security and privacy online. Prior experimental work has shown that social media are effective platforms for encouraging security-enhancing behavior. Through an analysis of historical social media logs of 38 participants containing almost 200,000 social media posts, we study the extent to which participants talked about security and privacy on social media platforms, specifically Facebook and Twitter. We found that interactions with posts that feature content relevant to security and privacy made up less than 0.09% of all interactions we observed. A thematic analysis of the security- and privacy-related posts that participants interacted with revealed that such posts very rarely discussed security and privacy constructively, instead often joking about security practices or encouraging undesirable behavior. Based on the overall findings from this thematic analysis, we develop and present a taxonomy of how security and privacy may be typically discussed on social networks, which is useful for constructing helpful security and privacy advice or for identifying advice that may have an undesirable impact. Our findings, though based on a fraction of the population of social media users, suggest that while social networks may be effective in influencing security behavior, there may not be enough substantial or useful discussions of security and privacy to encourage better security behaviors in practice and on a larger scale. Our findings highlight the importance of increasing the prevalence of constructive security and privacy advice on online social media in order to encourage widespread adoption of healthy security practices.

Abstract Over half of all visits to websites now take place in a mobile browser, yet the majority of web privacy studies take the vantage point of desktop browsers, use emulated mobile browsers, or focus on just a single mobile browser instead. In this paper, we present a comprehensive web-tracking measurement study on mobile browsers and privacy-focused mobile browsers. Our study leverages a new web measurement infrastructure, OmniCrawl, which we develop to drive browsers on desktop computers and smartphones located on two continents. We capture web tracking measurements using 42 different non-emulated browsers simultaneously. We find that the third-party advertising and tracking ecosystem of mobile browsers is more similar to that of desktop browsers than previous findings suggested. We study privacy-focused browsers and find their protections differ significantly and in general are less for lower-ranked sites. Our findings also show that common methodological choices made by web measurement studies, such as the use of emulated mobile browsers and Selenium, can lead to website behavior that deviates from what actual users experience.

Motivated by the transformative impact of deep neural networks (DNNs) in various domains, researchers and anti-virus vendors have proposed DNNs for malware detection from raw bytes that do not require manual feature engineering. In this work, we propose an attack that interweaves binary-diversification techniques and optimization frameworks to mislead such DNNs while preserving the functionality of binaries. Unlike prior attacks, ours manipulates instructions that are a functional part of the binary, which makes it particularly challenging to defend against. We evaluated our attack against three DNNs in white- and black-box settings, and found that it often achieved success rates near 100%. Moreover, we found that our attack can fool some commercial anti-viruses, in certain cases with a success rate of 85%. We explored several defenses, both new and old, and identified some that can foil over 80% of our evasion attempts. However, these defenses may still be susceptible to evasion by attacks, and so we advocate for augmenting malware-detection systems with methods that do not rely on machine learning.

Client-side cross-site scripting (DOM XSS) vulnerabilities in web applications are common, hard to identify, and difficult to prevent. Taint tracking is the most promising approach for detecting DOM XSS with high precision and recall, but is too computationally expensive for many practical uses.

Smartphone-based contact-tracing apps are a promising solution to help scale up the conventional contact-tracing process. However, low adoption rates have become a major issue that prevents these apps from achieving their full potential. In this paper, we present a national-scale survey experiment (N=1963) in the U.S. to investigate the effects of app design choices and individual differences on COVID-19 contact-tracing app adoption intentions. We found that individual differences such as prosocialness, COVID-19 risk perceptions, general privacy concerns, technology readiness, and demographic factors played a more important role than app design choices such as decentralized design vs. centralized design, location use, app providers, and the presentation of security risks. Certain app designs could exacerbate the different preferences in different sub-populations which may lead to an inequality of acceptance to certain app design choices (e.g., developed by state health authorities vs. a large tech company) among different groups of people (e.g., people living in rural areas vs. people living in urban areas). Our mediation analysis showed that one’s perception of the public health benefits offered by the app and the adoption willingness of other people had a larger effect in explaining the observed effects of app design choices and individual differences than one’s perception of the app’s security and privacy risks. With these findings, we discuss practical implications on the design, marketing, and deployment of COVID-19 contact-tracing apps in the U.S.

Abstract Recent research and articles in popular press have raised concerns about the privacy risks that smart home devices can create for incidental users—people who encounter smart home devices that are owned, controlled, and configured by someone else. In this work, we present the results of a user-centered investigation that explores incidental users’ experiences and the tensions that arise between device owners and incidental users. We conducted five focus group sessions through which we identified specific contexts in which someone might encounter other people’s smart home devices and the main concerns device owners and incidental users have in such situations. We used these findings to inform the design of a survey instrument, which we deployed to a demographically representative sample of 386 adults in the United States. Through this survey, we can better understand which contexts and concerns are most bothersome and how often device owners are willing to accommodate incidental users’ privacy preferences. We found some surprising trends in terms of what people are most worried about and what actions they are willing to take. For example, while participants who did not own devices themselves were often uncomfortable imagining them in their own homes, they were not as concerned about being affected by such devices in homes that they entered as part of their jobs. Participants showed interest in privacy solutions that might have a technical implementation component, but also frequently envisioned an open dialogue between incidental users and device owners to negotiate privacy accommodations.

Abstract Cameras are everywhere, and are increasingly coupled with video analytics software that can identify our face, track our mood, recognize what we are doing, and more. We present the results of a 10-day in-situ study designed to understand how people feel about these capabilities, looking both at the extent to which they expect to encounter them as part of their everyday activities and at how comfortable they are with the presence of such technologies across a range of realistic scenarios. Results indicate that while some widespread deployments are expected by many (e.g., surveillance in public spaces), others are not, with some making people feel particularly uncomfortable. Our results further show that individuals’ privacy preferences and expectations are complicated and vary with a number of factors such as the purpose for which footage is captured and analyzed, the particular venue where it is captured, and whom it is shared with. Finally, we discuss the implications of people’s rich and diverse preferences on opt-in or opt-out rights for the collection and use (including sharing) of data associated with these video analytics scenarios as mandated by regulations. Because of the user burden associated with the large number of privacy decisions people could be faced with, we discuss how new types of privacy assistants could possibly be configured to help people manage these decisions.

Multiple mechanisms exist to encourage users to create stronger passwords, including minimum-length and character-class requirements, prohibiting blocklisted passwords, and giving feedback on the strength of candidate passwords. Despite much research, there is little definitive, scientific guidance on how these mechanisms should be combined and configured to best effect. Through two online experiments, we evaluated combinations of minimum-length and character-class requirements, blocklists, and a minimum-strength requirement that requires passwords to exceed a strength threshold according to neural-network-driven password-strength estimates.

A long-standing technique to interfere with theft of sensitive data by its intended users is permitting these insiders only remote access to the data via a thin client. Even allowing only remote access is inadequate, however, to counter an insider willing to reconstruct the data from the graphical output, in the limit by photographing the data on-screen and applying automatic character recognition to these photographs offline. In this paper we propose and evaluate a system, called Snowman, that accurately monitors the amount of sensitive data output to a client. To conduct this monitoring without slowing the interactive user session, leakage is concurrently tracked in a replica of the application execution. This, in turn, introduces a key technical challenge that Snowman solves, namely identically replicating execution of an unmodified Linux binary while also performing efficient multi-label taint-tracking on it. We show through empirical measurements with a word processor, a spreadsheet program, and a code editor that Snowman induces little overhead on interactive user sessions and easily differentiates data-access patterns induced by normal usage and sufficiently aggressive data theft with reasonable responsiveness.

Understanding users' perceptions of suspected computer-security problems can help us tailor technology to better protect users. To this end, we conducted a field study of users' perceptions using 189,272 problem descriptions sent to the customer-support desk of a large anti-virus vendor from 2015 to 2018. Using qualitative methods, we analyzed 650 problem descriptions to study the security issues users faced and the symptoms that led users to their own diagnoses. Subsequently, we investigated to what extent and for what types of issues user diagnoses matched those of experts. We found, for example, that users and experts were likely to agree for most issues, but not for attacks (e.g., malware infections), for which they agreed only in 44% of the cases. Our findings inform several user-security improvements, including how to automate interactions with users to resolve issues and to better communicate issues to users.

Images perturbed subtly to be misclassified by neural networks, calledadversarial examples, have emerged as a technically deep challenge and an important concern for several application domains. Most research on adversarial examples takes as its only constraint that the perturbed images are similar to the originals. However, real-world application of these ideas often requires the examples to satisfy additional objectives, which are typically enforced through custom modifications of the perturbation process. In this article, we proposeadversarial generative nets(AGNs), a general methodology to train ageneratorneural network to emit adversarial examples satisfying desired objectives. We demonstrate the ability of AGNs to accommodate a wide range of objectives, including imprecise ones difficult to model, in two application domains. In particular, we demonstratephysicaladversarial examples—eyeglass frames designed to fool face recognition—with better robustness, inconspicuousness, and scalability than previous approaches, as well as a new attack to fool a handwritten-digit classifier.

Despite the additional protection it affords, two-factor authentication (2FA) adoption reportedly remains low. To better understand 2FA adoption and its barriers, we observed the deployment of a 2FA system at Carnegie Mellon University (CMU). We explore user behaviors and opinions around adoption, surrounding a mandatory adoption deadline. Our results show that (a) 2FA adopters found it annoying, but fairly easy to use, and believed it made their accounts more secure; (b) experience with CMU Duo often led to positive perceptions, sometimes translating into 2FA adoption for other accounts; and, (c) the differences between users required to adopt 2FA and those who adopted voluntarily are smaller than expected. We also explore the relationship between different usage patterns and perceived usability, and identify user misconceptions, insecure practices, and design issues. We conclude with recommendations for large-scale 2FA deployments to maximize adoption, focusing on implementation design, use of adoption mandates, and strategic messaging.

As increasingly many Internet-of-Things (IoT) devices collect personal data, users face more privacy decisions. Personal privacy assistants can provide social cues and help users make informed decisions by presenting information about how others have decided in similar cases. To better understand which social cues are relevant and whose recommendations users are more likely to follow, we presented 1000 online participants with nine IoT data-collection scenarios. Some participants were told the percentage of experts or friends who allowed data collection in each scenario, while other participants were provided no social cue. At the conclusion of each scenario, participants were asked whether they would allow the described data collection. Our results help explain under what circumstances users are more or less likely to be swayed by the reported behavior of others in similar scenarios. For example, our results indicate that when friends denied data collection, our participants were more influenced than when friends allowed data collection. On the other hand, participants were more influenced by experts when they allowed data collection. We also observed that influence could get stronger or wear off when participants were exposed to a sequence of scenarios. For example, when experts and friends repeatedly allowed data collection in scenarios with clear risk or denied it in scenarios with clear benefits, participants were less likely to be influenced by them in subsequent scenarios.

Text passwords---a frequent vector for account compromise, yet still ubiquitous---have been studied for decades by researchers attempting to determine how to coerce users to create passwords that are hard for attackers to guess but still easy for users to type and memorize. Most studies examine one password or a small number of passwords per user, and studies often rely on passwords created solely for the purpose of the study or on passwords protecting low-value accounts. These limitations severely constrain our understanding of password security in practice, including the extent and nature of password reuse, password behaviors specific to categories of accounts (e.g., financial websites), and the effect of password managers and other privacy tools.

Many authentication schemes ask users to manually compare compact representations of cryptographic keys, known as fingerprints. If the fingerprints do not match, that may signal a man-in-the-middle attack. An adversary performing an attack may use a fingerprint that is similar to the target fingerprint, but not an exact match, to try to fool inattentive users. Fingerprint representations should thus be both usable and secure. We tested the usability and security of eight fingerprint representations under different configurations. In a 661-participant between-subjects experiment, participants compared fingerprints under realistic conditions and were subjected to a simulated attack. The best configuration allowed attacks to succeed 6% of the time; the worst 72%. We find the seemingly effective compare-and-select approach performs poorly for key fingerprints and that graphical fingerprint representations, while intuitive and fast, vary in performance. We identify some fingerprint representations as particularly promising.

Despite their ubiquity, many password meters provide inaccurate strength estimates. Furthermore, they do not explain to users what is wrong with their password or how to improve it. We describe the development and evaluation of a data-driven password meter that provides accurate strength measurement and actionable, detailed feedback to users. This meter combines neural networks and numerous carefully combined heuristics to score passwords and generate data-driven text feedback about the user's password. We describe the meter's iterative development and final design. We detail the security and usability impact of the meter's design dimensions, examined through a 4,509-participant online study. Under the more common password-composition policy we tested, we found that the data-driven meter with detailed feedback led users to create more secure, and no less memorable, passwords than a meter with only a bar as a strength indicator.

The use of end-user programming, such as if-this-then-that (IFTTT), is becoming increasingly common. Services like IFTTT allow users to easily create new functionality by connecting arbitrary Internet-of-Things (IoT) devices and online services using simple if-then rules, commonly known as recipes. However, such convenience at times comes at the cost of security and privacy risks for end users. To gain an in-depth understanding of the potential security and privacy risks, we build an information-flow model to analyze how often IFTTT recipes involve potential integrity or secrecy violations. Our analysis finds that around 50% of the 19,323 unique recipes we examined are potentially unsafe, as they contain a secrecy violation, an integrity violation, or both. We next categorize the types of harm that these potentially unsafe recipes can cause to users. After manually examining a random selection of potentially unsafe recipes, we find that recipes can not only lead to harms such as personal embarrassment but can also be exploited by an attacker, e.g., to distribute malware or carry out denial-of-service attacks. The use of IoT devices and services like IFTTT is expected only to grow in the near future; our analysis suggests users need to be both informed about and protected from these emerging threats to which they could be unwittingly exposing themselves.

The special issue of the Journal of Computer Security contained three papers that represent the latest developments in distributed systems and programming languages for information flow control. An additional two articles were commissioned for this special issue that do the same for specialist hardware architectures and virtualisation platforms. Each paper was specially solicited by approaching authors of major papers on verified information flow security that had recently appeared in the top conference venues. The publishers carefully selected work that focused on practicality, and presented feature-rich systems or platforms with strong information flow guarantees, that have remained under active development, while ensuring we covered the broadest portion of the traditional hardware and software stack.

Machine learning is enabling a myriad innovations, including new algorithms for cancer diagnosis and self-driving cars. The broad use of machine learning makes it important to understand the extent to which machine-learning algorithms are subject to attack, particularly when used in applications where physical security or safety is at risk.

Recent research has improved our understanding of how to create strong, memorable text passwords. However, this research has generally been in the context of desktops and laptops, while users are increasingly creating and entering passwords on mobile devices. In this paper we study whether recent password guidance carries over to the mobile setting. We compare the strength and usability of passwords created and used on mobile devices with those created and used on desktops and laptops, while varying password policy requirements and input methods. We find that creating passwords on mobile devices takes significantly longer and is more error prone and frustrating. Passwords created on mobile devices are also weaker, but only against attackers who can make more than 10^13 guesses. We find that the effects of password policies differ between the desktop and mobile environments, and suggest ways to ease password entry for mobile users.

People share personal content online with varied audiences, as part of tasks ranging from conversational-style content sharing to collaborative activities. We use an interview- and diary-based study to explore: 1) what factors impact channel choice for sharing with particular audiences; and 2) what behavioral patterns emerge from the ability to combine or switch between channels. We find that in the context of different tasks, participants match channel features to selective-sharing and other task-based needs, shaped by recipient attributes and communication dynamics. Participants also combine multiple channels to create composite sharing features or reach broader audiences when one channel is insufficient. We discuss design implications of these channel dynamics.

Although many users create predictable passwords, the extent to which users realize these passwords are predictable is not well understood. We investigate the relationship between users' perceptions of the strength of specific passwords and their actual strength. In this 165-participant online study, we ask participants to rate the comparative security of carefully juxtaposed pairs of passwords, as well as the security and memorability of both existing passwords and common password-creation strategies. Participants had serious misconceptions about the impact of basing passwords on common phrases and including digits and keyboard patterns in passwords. However, in most other cases, participants' perceptions of what characteristics make a password secure were consistent with the performance of current password-cracking tools. We find large variance in participants' understanding of how passwords may be attacked, potentially explaining why users nonetheless make predictable passwords. We conclude with design directions for helping users make better passwords.

Abstract Online trackers compile profiles on users for targeting ads, customizing websites, and selling users’ information. In this paper, we report on the first detailed study of the perceived benefits and risks of tracking-and the reasons behind them-conducted in the context of users’ own browsing histories. Prior work has studied this in the abstract; in contrast, we collected browsing histories from and interviewed 35 people about the perceived benefits and risks of online tracking in the context of their own browsing behavior. We find that many users want more control over tracking and think that controlled tracking has benefits, but are unwilling to put in the effort to control tracking or distrust current tools. We confirm previous findings that users’ general attitudes about tracking are often at odds with their comfort in specific situations. We also identify specific situational factors that contribute to users’ preferences about online tracking and explore how and why. Finally, we examine a sample of popular tools for controlling tracking and show that they only partially address the situational factors driving users’ preferences.We suggest opportunities to improve such tools, and explore the use of a classifier to automatically determine whether a user would be comfortable with tracking on a particular page visit; our results suggest this is a promising direction for future work.

Password-composition policies are the result of service providers becoming increasingly concerned about the security of online accounts. These policies restrict the space of user-created passwords to preclude easily guessed passwords and thus make passwords more difficult for attackers to guess. However, many users struggle to create and recall their passwords under strict password-composition policies, for example, ones that require passwords to have at least eight characters with multiple character classes and a dictionary check. Recent research showed that a promising alternative was to focus policy requirements on password length instead of on complexity. In this work, we examine 15 password policies, many focusing on length requirements. In doing so, we contribute the first thorough examination of policies requiring longer passwords. We conducted two online studies with over 20,000 participants, and collected both usability and password-strength data. Our findings indicate that password strength and password usability are not necessarily inversely correlated: policies that lead to stronger passwords do not always reduce usability. We identify policies that are both more usable and more secure than commonly used policies that emphasize complexity rather than length requirements. We also provide practical recommendations for service providers who want their users to have strong yet usable passwords.

Users often struggle to create passwords under strict requirements. To make this process easier, some providers present real-time feedback during password creation, indicating which requirements are not yet met. Other providers guide users through a multi-step password-creation process. Our 6,435-participant online study examines how feedback and guidance affect password security and usability. We find that real-time password-creation feedback can help users create strong passwords with fewer errors. We also find that although guiding participants through a three-step password-creation process can make creation easier, it may result in weaker passwords. Our results suggest that service providers should present password requirements with feedback to increase usability. However, the presentation of feedback and guidance must be carefully considered, since identical requirements can have different security and usability effects depending on presentation.

This paper presents a formal framework for run-time enforcement mechanisms, or monitors, based on probabilistic input/output automata [Task-structured probabilistic I/O automata, Technical Report MIT-CSAIL-TR-2006-060, 2006; Proceedings of the 8th International Workshop on Discrete Event Systems , 2006, pp. 207–214], which allows for the modeling of complex and interactive systems. We associate with each trace of a monitored system (i.e., a monitor interposed between a system and an environment) a probability and a real number that represents the cost that the actions appearing on the trace incur on the monitored system. This allows us to calculate the probabilistic (expected) cost of the monitor and the monitored system, which we use to classify monitors, not only in the typical sense, as sound and transparent [ ACM Transactions on Information and System Security 12 (3) (2009), 1–41], but also at a more fine-grained level, as cost-optimal or cost-efficient. We show how a cost-optimal monitor can be built using information about cost and the probabilistic future behavior of the system and the environment, showing how deeper knowledge of a system can lead to construction of more efficient security mechanisms.

To encourage strong passwords, system administrators employ password-composition policies, such as a traditional policy requiring that passwords have at least 8 characters from 4 character classes and pass a dictionary check. Recent research has suggested, however, that policies requiring longer passwords with fewer additional requirements can be more usable and in some cases more secure than this traditional policy. To explore long passwords in more detail, we conducted an online experiment with 8,143 participants. Using a cracking algorithm modified for longer passwords, we evaluate eight policies across a variety of metrics for strength and usability. Among the longer policies, we discover new evidence for a security/usability tradeoff, with none being strictly better than another on both dimensions. However, several policies are both more usable and more secure that the traditional policy we tested. Our analyses additionally reveal common patterns and strings found in cracked passwords. We discuss how system administrators can use these results to improve password-composition policies.

We are seeing a significant shift in the types and characteristics of computing devices that are commonly used. Today, more smartphones are sold than personal computers. An area of rapid growth are also cloud systems; and our everyday lives are invaded by sensors like smart meters and electronic tickets. The days when most computing resources were managed directly by a computer's operating system are over---data and computation is distributed, and devices are typically always connected via the Internet.

Despite considerable research on passwords, empirical studies of password strength have been limited by lack of access to plaintext passwords, small data sets, and password sets specifically collected for a research study or from low-value accounts. Properties of passwords used for high-value accounts thus remain poorly understood.

Much of the debate surrounding online behavioral advertising (OBA) has centered on how to provide users with notice and choice. An important element left unexplored is how advertising companies' privacy practices affect users' attitudes toward data sharing. We present the results of a 2,912-participant online study investigating how facets of privacy practices---data retention, access to collected data, and scope of use---affect users' willingness to allow the collection of behavioral data. We asked participants to visit a health website, explained OBA to them, and outlined policies governing data collection for OBA purposes. These policies varied by condition. We then asked participants about their willingness to permit the collection of 30 types of information. We identified classes of information that most participants would not share, as well as classes that nearly half of participants would share. More restrictive data-retention and scope-of-use policies increased participants' willingness to allow data collection. In contrast, whether the data was collected on a well-known site and whether users could review and modify their data had minimal impact. We discuss public policy implications and improvements to user interfaces to align with users' privacy preferences.

Users often have rich and complex photo-sharing preferences, but properly configuring access control can be difficult and time-consuming. In an 18-participant laboratory study, we explore whether the keywords and captions with which users tag their photos can be used to help users more intuitively create and maintain access-control policies. We find that (a) tags created for organizational purposes can be repurposed to create efficient and reasonably accurate access-control rules; (b) users tagging with access control in mind develop coherent strategies that lead to significantly more accurate rules than those associated with organizational tags alone; and (c) participants can understand and actively engage with the concept of tag-based access control.

Accesses that are not permitted by implemented policy but that share similarities with accesses that have been allowed, may be indicative of access-control policy misconfigurations. Identifying such misconfigurations allows administrators to resolve them before they interfere with the use of the system. We improve upon prior work in identifying such misconfigurations in two main ways. First, we develop a new methodology for evaluating misconfiguration prediction algorithms and applying them to real systems. We show that previous evaluations can substantially overestimate the benefits of using such algorithms in practice, owing to their tendency to reward predictions that can be deduced to be redundant. We also show, however, that these and other deductions can be harnessed to substantially recover the benefits of prediction. Second, we propose an approach that significantly simplifies the use of misconfiguration prediction algorithms. We remove the need to hand-tune (and empirically determine the effects of) various parameters, and instead replace them with a single, intuitive tuning parameter. We show empirically that this approach is generally competitive in terms of benefit and accuracy with algorithms that require hand-tuned parameters.

We take a detailed look at how users, while focusing on non-permission tasks, notice and fix access-control permission errors depending on where the access-control policy is spatially located on a photo-sharing website. The access-control policy was placed on an online photo-sharing website under the photo or album, on the sidebar, or on a separate settings page. We find that placing the access-control policy directly under photos and album thumbnails improves participants' ability to notice errors in their access-control settings without negatively impacting non-access-control tasks.

Users tend to create passwords that are easy to guess, while system-assigned passwords tend to be hard to remember. Passphrases, space-delimited sets of natural language words, have been suggested as both secure and usable for decades. In a 1,476-participant online study, we explored the usability of 3- and 4-word system-assigned passphrases in comparison to system-assigned passwords composed of 5 to 6 random characters, and 8-character system-assigned pronounceable passwords. Contrary to expectations, system-assigned passphrases performed similarly to system-assigned passwords of similar entropy across the usability metrics we examined. Passphrases and passwords were forgotten at similar rates, led to similar levels of user difficulty and annoyance, and were both written down by a majority of participants. However, passphrases took significantly longer for participants to enter, and appear to require error-correction to counteract entry mistakes. Passphrase usability did not seem to increase when we shrunk the dictionary from which words were chosen, reduced the number of words in a passphrase, or allowed users to change the order of words.

Text-based passwords remain the dominant authentication method in computer systems, despite significant advancement in attackers' capabilities to perform password cracking. In response to this threat, password composition policies have grown increasingly complex. However, there is insufficient research defining metrics to characterize password strength and using them to evaluate password-composition policies. In this paper, we analyze 12,000 passwords collected under seven composition policies via an online study. We develop an efficient distributed method for calculating how effectively several heuristic password-guessing algorithms guess passwords. Leveraging this method, we investigate (a) the resistance of passwords created under different conditions to guessing, (b) the performance of guessing algorithms under different training sets, (c) the relationship between passwords explicitly created under a given composition policy and other passwords that happen to meet the same requirements, and (d) the relationship between guess ability, as measured with password-cracking algorithms, and entropy estimates. Our findings advance understanding of both password-composition policies and metrics for quantifying password security.

As the capabilities of smartphones increase, users are beginning to rely on these mobile and ubiquitous platforms to perform more tasks. In addition to traditional computing tasks, people are beginning to use smartphones to interact with people they meet. Often this interaction begins with an exchange, e.g., of cryptographic keys. Hence, a number of protocols have been developed to facilitate this exchange. Unfortunately, those protocols that provide strong security guarantees often suffer from usability problems, and easy-to-use protocols may lack the desired security guarantees.

Text-based passwords are the most common mechanism for authenticating humans to computer systems. To prevent users from picking passwords that are too easy for an adversary to guess, system administrators adopt password-composition policies (e.g., requiring passwords to contain symbols and numbers). Unfortunately, little is known about the relationship between password-composition policies and the strength of the resulting passwords, or about the behavior of users (e.g., writing down passwords) in response to different policies. We present a large-scale study that investigates password strength, user behavior, and user sentiment across four password-composition policies. We characterize the predictability of passwords by calculating their entropy, and find that a number of commonly held beliefs about password composition and strength are inaccurate. We correlate our results with user behavior and sentiment to produce several recommendations for password-composition policies that result in strong passwords without unduly burdening users.

As users store and share more digital content at home, access control becomes increasingly important. One promising approach for helping non-expert users create accurate access policies is reactive policy creation, in which users can update their policy dynamically in response to access requests that would not otherwise succeed. An earlier study suggested reactive policy creation might be a good fit for file access control at home. To test this, we conducted an experience-sampling study in which participants used a simulated reactive access-control system for a week. Our results bolster the case for reactive policy creation as one mode by which home users specify access-control policy. We found both quantitative and qualitative evidence of dynamic, situational policies that are hard to implement using traditional models but that reactive policy creation can facilitate. While we found some clear disadvantages to the reactive model, they do not seem insurmountable.

In access-control systems, policy rules conflict when they prescribe different decisions (allow or deny) for the same access. We present the results of a user study that demonstrates the significant impact of conflict-resolution method on policy-authoring usability. In our study of 54 participants, varying the conflict-resolution method yielded statistically significant differences in accuracy in five of the six tasks we tested, including differences in accuracy rates of up to 78%. Our results suggest that a conflict-resolution method favoring rules of smaller scope over rules of larger scope is more usable than the Microsoft Windows operating system's method of favoring deny rules over allow rules. Perhaps more importantly, our results demonstrate that even seemingly small changes to a system's semantics can fundamentally affect the system's usability in ways that are beyond the power of user interfaces to correct.

Access-control policy misconfigurations that cause requests to be erroneously denied can result in wasted time, user frustration, and, in the context of particular applications (e.g., health care), very severe consequences. In this article we apply association rule mining to the history of accesses to predict changes to access-control policies that are likely to be consistent with users' intentions, so that these changes can be instituted in advance of misconfigurations interfering with legitimate accesses. Instituting these changes requires the consent of the appropriate administrator, of course, and so a primary contribution of our work is how to automatically determine from whom to seek consent and how to minimize the costs of doing so. We show using data from a deployed access-control system that our methods can reduce the number of accesses that would have incurred costly time-of-access delays by 43%, and can correctly predict 58% of the intended policy. These gains are achieved without impacting the total amount of time users spend interacting with the system.

As digital content becomes more prevalent in the home, non-technical users are increasingly interested in sharing that content with others and accessing it from multiple devices. Not much is known about how these users think about controlling access to this data. To better understand this, we conducted semi-structured, in-situ interviews with 33 users in 15 households. We found that users create ad-hoc access-control mechanisms that do not always work; that their ideal policies are complex and multi-dimensional; that a priori policy specification is often insufficient; and that people's mental models of access control and security are often misaligned with current systems. We detail these findings and present a set of associated guidelines for designing usable access-control systems for the home environment.

Authorization logics allow concise specification of flexible access-control policies, and are the basis for logic-based access-control systems. In such systems, resource owners issue credentials to specify policies, and the consequences of these policies are derived using logical inference rules. Proofs in authorization logics can serve as capabilities for gaining access to resources. Because a proof is derived from a set of credentials possibly issued by different parties, the issuer of a specific credential may not be aware of all the proofs that her credential may make possible. From this credential issuer's standpoint, the policy expressed in her credential may thus have unexpected consequences. To solve this general problem, we propose a system in which credentials can specify constraints on how they are to be used. We show how to modularly extend wellstudied authorization logics to support the specification and enforcement of such constraints. A novelty of our design is that we allow the constraints to be arbitrary well-behaved functions over authorization proofs. Since all the information about an access is contained in the proofs, this makes it possible to express many interesting constraints. We study the formal properties of such a system, and give examples of constraints.

No abstract available.

Text-based passwords are still the most commonly used authentication mechanism in information systems. We took advantage of a unique opportunity presented by a significant change in the Carnegie Mellon University (CMU) computing services password policy that required users to change their passwords. Through our survey of 470 CMU computer users, we collected data about behaviors and practices related to the use and creation of passwords. We also captured users' opinions about the new, stronger policy requirements. Our analysis shows that, although most of the users were annoyed by the need to create a complex password, they believe that they are now more secure. Furthermore, we perform an entropy analysis and discuss how our findings relate to NIST recommendations for creating a password policy. We also examine how users answer specific questions related to their passwords. Our results can be helpful in designing better password policies that consider not only technical aspects of specific policy rules, but also users' behavior in response to those rules.

In this work we ask the question: what are the challenges of managing a physical or file system access-control policy for a large organization? To answer the question, we conducted a series of interviews with thirteen administrators who manage access-control policy for either a file system or a physical space. Based on these interviews we identified three sets of real-world requirements that are either ignored or inadequately addressed by technology: 1) policies are made/implemented by multiple people; 2) policy makers are distinct from policy implementers; and 3) access-control systems don't always have the capability to implement the desired policy. We present our interview results and propose several possible solutions to address the observed issues.

A number of research systems have demonstrated the benefits of accompanying each request with a machine-checkable proof that the request complies with access-control policy - a technique called proof-carrying authorization. Numerous authorization logics have been proposed as vehicles by which these proofs can be expressed and checked. A challenge in building such systems is how to allow delegation between institutions that use different authorization logics. Instead of trying to develop the authorization logic that all institutions should use, we propose a framework for interfacing different, mutually incompatible authorization logics. Our framework provides a very small set of primitives that defines an interface for communication between different logics without imposing any fundamental constraints on their design or nature. We illustrate by example that a variety of different logics can communicate over this interface, and show formally that supporting the interface does not impinge on the integrity of each individual logic. We also describe an architecture for constructing authorization proofs that contain components from different logics and report on the performance of a prototype proof checker.

A common mechanism for ensuring that software behaves securely is to monitor programs at run time and check that they dynamically adhere to constraints specified by a security policy. Whenever a program monitor detects that untrusted software is attempting to execute a dangerous action, it takes remedial steps to ensure that only safe code actually gets executed. This article improves our understanding of the space of policies enforceable by monitoring the run-time behaviors of programs. We begin by building a formal framework for analyzing policy enforcement: we precisely define policies, monitors, and enforcement. This framework allows us to prove that monitors enforce an interesting set of policies that we call the infinite renewal properties. We show how to construct a program monitor that provably enforces any reasonable infinite renewal property. We also show that the set of infinite renewal properties includes some nonsafety policies, that is, that monitors can enforce some nonsafety (including some purely liveness) policies. Finally, we demonstrate concrete examples of nonsafety policies enforceable by practical run-time monitors.

Program monitors enforce security policies by interposing themselves into the control flow of untrusted software whenever that software attempts to execute security-relevant actions. At the point of interposition, a monitor has authority to permit or deny (perhaps conditionally) the untrusted software's attempted action. Program monitors are common security enforcement mechanisms and integral parts of operating systems, virtual machines, firewalls, network auditors, and antivirus and antispyware tools. Unfortunately, the runtime policies we require program monitors to enforce grow more complex, both as the monitored software is given new capabilities and as policies are refined in response to attacks and user feedback. We propose dealing with policy complexity by organizing policies in such a way as to make them composable, so that complex policies can be specified more simply as compositions of smaller subpolicy modules. We present a fully implemented language and system called Polymer that allows security engineers to specify and enforce composable policies on Java applications. We formalize the central workings of Polymer by defining an unambiguous semantics for our language. Using this formalization, we state and prove an uncircumventability theorem which guarantees that monitors will intercept all security-relevant actions of untrusted software.

Significant effort has been invested in developing expressive and flexible access-control languages and systems. However, little has been done to evaluate these systems in practical situations with real users, and few attempts have been made to discover and analyze the access-control policies that users actually want to implement. We report on a user study in which we derive the ideal access policies desired by a group of users for physical security in an office environment. We compare these ideal policies to the policies the users actually implemented with keys and with a smartphone-based distributed access-control system. We develop a methodology that allows us to show quantitatively that the smartphone system allowed our users to implement their ideal policies more accurately and securely than they could with keys, and we describe where each system fell short.

We introduce the Expandable Grid, a novel interaction technique for creating, editing, and viewing many types of security policies. Security policies, such as file permissions policies, have traditionally been displayed and edited in user interfaces based on a list of rules, each of which can only be viewed or edited in isolation. These list-of-rules interfaces cause problems for users when multiple rules interact, because the interfaces have no means of conveying the interactions amongst rules to users. Instead, users are left to figure out these rule interactions themselves. An Expandable Grid is an interactive matrix visualization designed to address the problems that list-of-rules interfaces have in conveying policies to users. This paper describes the Expandable Grid concept, shows a system using an Expandable Grid for setting file permissions in the Microsoft Windows XP operating system, and gives results of a user study involving 36 participants in which the Expandable Grid approach vastly outperformed the native Windows XP file-permissions interface on a broad range of policy-authoring tasks.

Access-control policy misconfigurations that cause requests to be erroneously denied can result in wasted time, user frustration and, in the context of particular applications (e.g., health care), very severe consequences. In this paper we apply association rule mining to the history of accesses to predict changes to access-control policies that are likely to be consistent with users' intentions, so that these changes can be instituted in advance of misconfigurations interfering with legitimate accesses. Instituting these changes requires consent of the appropriate administrator, of course, and so a primary contribution of our work is to automatically determine from whom to seek consent and to minimize the costs of doing so. We show using data from a deployed access-control system that our methods can reduce the number of accesses that would have incurred costly time-of-access delays by 44%, and can correctly predict 58% of the intended policy. These gains are achieved without increasing the total amount of time users spend interacting with the system.

Grey is a smartphone-based system by which a user can exercise her authority to gain access to rooms in our university building, and by which she can delegate that authority to other users. We present findings from a trial of Grey, with emphasis on how common usability principles manifest themselves in a smartphone-based security application. In particular, we demonstrate aspects of the system that gave rise to failures, misunderstandings, misperceptions, and unintended uses; network effects and new flexibility enabled by Grey; and the implications of these for user behavior. We argue that the manner in which usability principles emerged in the context of Grey can inform the design of other such applications.

We introduce a language and system that supports definition and composition of complex run-time security policies for Java applications. Our policies are comprised of two sorts of methods. The first is query methods that are called whenever an untrusted application tries to execute a security-sensitive action. A query method returns a suggestion indicating how the security-sensitive action should be handled. The second sort of methods are those that perform state updates as the policy's suggestions are followed.The structure of our policies facilitates composition, as policies can query other policies for suggestions. In order to give programmers control over policy composition, we have designed the system so that policies, suggestions, and application events are all first-class objects that a higher-order policy may manipulate. We show how to use these programming features by developing a library of policy combinators.Our system is fully implemented, and we have defined a formal semantics for an idealized subset of the language containing all of the key features. We demonstrate the effectiveness of our system by implementing a large-scale security policy for an email client.

We present a distributed algorithm for assembling a proof that a request satisfies an access-control policy expressed in a formal logic, in the tradition of Lampson et al. (1992). We show analytically that our distributed proof-generation algorithm succeeds in assembling a proof whenever a centralized prover utilizing remote certificate retrieval would do so. In addition, we show empirically that our algorithm outperforms centralized approaches in various measures of performance and usability notably the number of remote requests and the number of user interruptions. We show that when combined with additional optimizations including caching and automatic tactic generation, which we introduce here, our algorithm retains its advantage, while achieving practical performance. Finally, we briefly describe the utilization of these algorithms as the basis for an access-control framework being deployed for use at our institution.

Abstract We present a new module system for Java that improves upon many of the deficiencies of the Java package system and gives the programmer more control over dynamic linking. Our module system provides explicit interfaces, multiple views of modules based on hierarchical nesting and more flexible name‐space management than the Java package system. Relationships between modules are explicitly specified in module description files. We provide more control over dynamic linking by allowing import statements in module description files to require that imported modules be annotated with certain properties, which we implement by digital signatures. Our module system is compatible enough with standard Java to be implemented as a source‐to‐source and bytecode‐to‐bytecode transformation wrapped around a standard Java compiler, using a standard Java virtual machine (JVM). Copyright © 2003 John Wiley & Sons, Ltd.