Answering end user security questions is challenging. While large language models (LLMs) like GPT, Llama, and Gemini are far from error-free, they have shown promise in answering a variety of questions outside of security. We qualitatively evaluated responses from three popular LLMs to 900 systematically collected security questions in the first such evaluation in the area of end user security. While LLMs demonstrate broad generalist “knowledge” of end user security information, there are patterns of errors and limitations across LLMs-including stale, inaccurate, and incomplete answers, as well as indirect or unresponsive communication styles-which negatively impact the user experience. Based on these patterns, we suggest directions for improved model development and recommend user strategies for interacting with LLMs when seeking assistance with security.

Digital financial services can introduce new digital-safety risks for users, particularly survivors of intimate partner financial abuse (IPFA). To offer improved support for such users, a comprehensive understanding of their support needs and the barriers they face to redress by financial institutions is essential. Drawing from a dataset of 2.7 million customer complaints, we implement a bespoke workflow that utilizes language-modeling techniques and expert human review to identify complaints describing IPFA. Our mixed-method analysis provides insight into the most common digital financial products involved in these attacks, and the barriers consumers report encountering when doing so. Our contributions are twofold; we offer the first human-labeled dataset for this overlooked harm and provide practical implications for technical practice, research, and design for better supporting and protecting survivors of IPFA.

Security incident management is of increasing importance given the growth in cybersecurity threats and regulatory attention. We provide what is, to the best of our knowledge, the first large-scale study of security incident management, consisting of a survey with more than 500 security incident management practitioners grounded by interviews with practitioners in 10 organizations, including companies and educational institutions. We find that some aspects of incident management workftows are consistent across organizations and incident management teams (e.g., collaboration and prioritization), but that the complexity of incident management generally grows with organization size, in that practitioners encounter a wider variety and volume of incidents, collaboration partners and tools. These variations suggest several ways in which tools can better support efficient incident management.

Privacy news influences end-user attitudes and behaviors as well as product and policy development, and so is an important data source for understanding privacy perceptions. We provide a largescale text mining of privacy news, focusing on patterns in sentiment and keywords. This is a challenging task given the lack of a privacy news repository and a ground truth for sentiment. Using high-precision data sets from two popular news sources in the U. S. and U. K., the New York Times and the Guardian, we find negative privacy news is far more common than positive. In addition, in the NYT, privacy news is more prominently reported than many world events involving significant human suffering. Our analysis provides a rich snapshot of this driver of privacy perceptions and demonstrates that news facilitates the systematization of privacy knowledge.

Website owners use web analytics tools to better understand their visitors for a range of purposes. However, there is limited understanding of how owners of non-transactional websites use and benefit from web analytics. Through semi-structured interviews (n=18) with non-transactional web analytics users we explore these uses and benefits. Participants tend to use web analytics to improve site design, by optimizing site structure, content, or technical specifications. However, participants also use web analytics to understand their audiences without a directed purpose, often for curiosity or entertainment. The design of web analytics tools should account for this full range of functionality.

We describe survey results from a representative sample of 1,075 U. S. social network users who use Facebook as their primary network. Our results show a strong association between low engagement and privacy concern. Specifically, users who report concerns around sharing control, comprehension of sharing practices or general Facebook privacy concern, also report consistently less time spent as well as less (self-reported) posting, commenting and "Like"ing of content. The limited evidence of other significant differences between engaged users and others suggests that privacy-related concerns may be an important gate to engagement. Indeed, privacy concern and network size are the only malleable attributes that we find to have significant association with engagement. We manually categorize the privacy concerns finding that many are nonspecific and not associated with negative personal experiences. Finally, we identify some education and utility issues associated with low social network activity, suggesting avenues for increasing engagement amongst current users.

The strong emotional reaction elicited by privacy issues is well documented (e.g., [12, 8]). The emotional aspect of privacy makes it difficult to evaluate privacy concern, and directly asking about a privacy issue may result in an emotional reaction and a biased response. This effect may be partly responsible for the dramatic privacy concern ratings coming from recent surveys, ratings that often seem to be at odds with user behavior. In this paper we propose indirect techniques for measuring content privacy concerns through surveys, thus hopefully diminishing any emotional response. We present a design for indirect surveys and test the design's use as (1) a means to measure relative privacy concerns across content types, (2) a tool for predicting unwillingness to share content (a possible indicator of privacy concern), and (3) a gauge for two underlying dimensions of privacy - content importance and the willingness to share content. Our evaluation consists of 3 surveys, taken by 200 users each, in which privacy is never asked about directly, but privacy warnings are issued with increasing escalation in the instructions and individual question-wording. We demonstrate that this escalation results in statistically and practically significant differences in responses to individual questions. In addition, we compare results against a direct privacy survey and show that rankings of privacy concerns are increasingly preserved as privacy language increases in the indirect surveys, thus indicating our mapping of the indirect questions to privacy ratings is accurately reflecting privacy concerns.

For privacy reasons, sensitive content may be revised before it is released. The revision often consists of redaction, that is, the blacking out of sensitive words and phrases. Redaction has the side effect of reducing the utility of the content, often so much that the content is no longer useful. Consequently, government agencies and others are increasingly exploring the revision of sensitive content as an alternative to redaction that preserves more content utility. We call this practice sanitization. In a sanitized document, names might be replaced with pseudonyms and sensitive attributes might be replaced with hypernyms. Sanitization adds to redaction the challenge of determining what words and phrases reduce the sensitivity of content. We have designed and developed a tool to assist users in sanitizing sensitive content. Our tool leverages the Web to automatically identify sensitive words and phrases and quickly evaluates revisions for sensitivity. The tool, however, does not identify all sensitive terms and mistakenly marks some innocuous terms as sensitive. This is unavoidable because of the difficulty of the underlying inference problem and is the main reason we have designed a sanitization assistant as opposed to a fully-automated tool. We have conducted a small study of our tool in which users sanitize biographies of celebrities to hide the celebrity's identity both both with and without our tool. The user study suggests that while the tool is very valuable in encouraging users to preserve content utility and can preserve privacy, this usefulness and apparent authoritativeness may lead to a slippery slope in which users neglect their own judgment in favor of the tool's.

Detecting inferences in documents is critical for ensuring privacy when sharing information. In this paper, we propose a refined and practical model of inference detection using a reference corpus. Our model is inspired by association rule mining: inferences are based on word co-occurrences. Using the model and taking the Web as the reference corpus, we can find inferences and measure their strength through web-mining algorithms that leverage search engines such as Google or Yahoo!.

As a result of recent changes in the policy governing internet content distribution, such as the institution of per listener royalties for internet radio broadcasters, Content Distributors (CDs) now have an incentive to underreport the size of their audience. Existing metering protocols protect only against the inflation of audience size. We introduce the first protocols for audience metering that protect against deflation attempts by CDs. The protocols trade-off the amount of additional information the CDs must distribute to facilitate metering with the amount of infrastructure required and are applicable to internet radio, web plagiarism, and software licence enforcement.

This paper proposes several new schemes which allow a sender to send encrypted messages to multiple recipients more efficiently (in terms of bandwidth and computation) than by using a standard encryption scheme. Most of the proposed schemes explore a new natural technique called randomness reuse. In order to analyze security of our constructions, we introduce a new notion of multirecipient encryption schemes (MRESs) and provide definitions of security for them. We finally show a way to avoid ad hoc analyses by providing a general test that can be applied to a standard encryption scheme to determine whether the associated randomness reusing MRES is secure. The results and applications cover both asymmetric and symmetric encryption.

When distributing digital content over a broadcast channel it's often necessary to revoke users whose access privileges have expired, thus preventing them from recovering the content. This works well when users make a conscious decision to leave the system or have misbehaved, but numerous cases exist in which the revocation is in error and users are consequently left with the often onerous burden of getting reinstated. We introduce a gradual form of revocation that we call service degradation that enables the content distributor to provide "cues" to the user in the form of degraded system performance. The cues alert the user to their impending revocation and allow them to take the necessary action to remain in the system. Our protocols build on techniques for broadcast encryption and spam-fighting to provide the appropriate form of service for this previously ignored class of users.

Access control can be used to ensure that database queries pertaining to sensitive information are not answered. This is not enough to prevent users from learning sensitive information though, because users can combine non-sensitive information to discover something sensitive. Inference control prevents users from obtaining sensitive information via such "inference channels", however, existing inference control techniques are not private - that is, they require the server to learn what queries the user is making in order to deny inference-enabling queries.

Consider a CIA agent who wants to authenticate herself to a server but does not want to reveal her CIA credentials unless the server is a genuine CIA outlet. Consider also that the CIA server does not want to reveal its CIA credentials to anyone but CIA agents - not even to other CIA servers. We first show how pairing-based cryptography can be used to implement such secret handshakes. We then propose a formal definition for secure secret handshakes, and prove that our pairing-based schemes are secure under the Bilinear Diffie-Hellman assumption. Our protocols support role-based group membership authentication, traceability, indistinguishability to eavesdroppers, unbounded collusion resistance, and forward repudiability. Our secret-handshake scheme can be implemented as a TLS cipher suite. We report on the performance of our preliminary Java implementation.

We apply results from algebraic coding theory to solve problems in cryptography, by using recent results on list decoding of error-correcting codes to efficiently find traitors who collude to create pirates. We produce schemes for which the traceability (TA) traitor tracing algorithm is very fast. We compare the TA and identifiable parent property (IPP) traitor tracing algorithms, and give evidence that when using an algebraic structure, the ability to trace traitors with the IPP algorithm implies the ability to trace with the TA algorithm. We also demonstrate that list decoding techniques can be used to find all possible pirate coalitions. Finally, we raise some related open questions about linear codes, and suggest uses for other decoding techniques in the presence of additional information about traitor behavior.

We address the problem of establishing a group key amongst a dynamic group of users over an unreliable, or lossy, network. We term our key distribution mechanisms self-healing because users are capable of recovering lost group keys on their own, without requesting additional transmissions from the group manager thus cutting back on network traffic, decreasing the load on the group manager and reducing the risk of user exposure through traffic analysis. A user must be a member both before and after the session in which a particular key is sent in order to be able to recover the key through self-healing. Binding the ability to recover keys to membership status enables the group manager to use short broadcasts to establish group keys, independent of the group size. In addition, the self-healing approach to key distribution is stateless, meaning that a group member who has been off-line for some time is able to recover new session keys immediately after coming back on-line.

We consider the authentication of digital streams over a lossy network. The overall approach taken is graph-based, as this yields simple methods for controlling overhead, delay, and the ability to authenticate, while serving to unify many previously known hash- and MAC-based techniques. The loss pattern of the network is defined probabilistically, allowing both bursty and random packet loss to be modeled. Our authentication schemes are customizable by the sender of the stream; that is, within reasonable constraints on the input parameters, we provide schemes that achieve the desired authentication probability while meeting the input upper bound on the overhead per packet. In addition, we demonstrate that some of the shortcomings of previously known schemes correspond to easily identifiable properties of a graph, and hence, may be more easily avoided by taking a graph-based approach to designing authentication schemes.

In order to protect copyrighted material, codes may be embedded in the content or codes may be associated with the keys used to recover the content. Codes can offer protection by providing some form of traceability (TA) for pirated data. Several researchers have studied different notions of TA and related concepts in previous years. "Strong" versions of TA allow at least one member of a coalition that constructs a "pirate decoder" to be traced. Weaker versions of this concept ensure that no coalition can "frame" a disjoint user or group of users. All these concepts can be formulated as codes having certain combinatorial properties. We study the relationships between the various notions, and we discuss equivalent formulations using structures such as perfect hash families. We use methods from combinatorics and coding theory to provide bounds (necessary conditions) and constructions (sufficient conditions) for the objects of interest.