Apple introduced app-based privacy labels in 2020 to improve apps' communication of their data practices. However, most developers appear to treat privacy labels as a ``set-once'' mechanism. To better understand the dynamics of this system, we first analyzed a four-year longitudinal dataset of Apple's Privacy Label. Next, we conducted an email survey of developers who have changed (or not changed) their privacy labels during this period, and finally, performed follow-up interviews with developers from each group. We find that only 51,364 apps (less than 6%) over this period have made any changes to their privacy labels, many of them changing their initial 'Do Not Collect' label to more refined classification. From the emails and interviews, the ``black box'' of third-party data practice may lead developers to underreport their app's data practices. Many developers reported that privacy labels are a valuable marketing tool for promoting their apps as privacy-friendly. Privacy labels may appear stable not necessarily because practices are stable, but because ambiguity encourages minimal or optimistic disclosure. To improve privacy label maintenance, we recommend enhanced transparency mechanisms for third-party libraries, stronger workflow integration, and platform support that guides developers and strengthens users' control.

This work presents Unbundle-Rewrite-Rebundle (URR), a system for detecting privacy-harming portions of bundled JavaScript code and rewriting that code at runtime to remove the privacy-harming behavior without breaking the surrounding code or overall application. URR is a novel solution to the problem of JavaScript bundles, where websites pre-compile multiple code units into a single file, making it impossible for content filters and ad-blockers to differentiate between desired and unwanted resources. Where traditional content filtering tools rely on URLs, URR analyzes the code at the AST level, and replaces harmful AST sub-trees with privacy-and-functionality maintaining alternatives.

Apple introduced privacy labels in Dec. 2020 as a way for developers to report the privacy behaviors of their apps. While Apple does not validate labels, they also require developers to provide a privacy policy, which offers an important comparison point. In this paper, we fine-tuned BERT-based language models to extract privacy policy features for 474,669 apps on the iOS App Store, comparing the output to the privacy labels. We identify discrepancies between the policies and the labels, particularly as they relate to data collected linked to users. We find that 228K apps' privacy policies may indicate data collection linked to users than what is reported in the privacy labels. More alarming, a large number (97%) of the apps with a Data Not Collected privacy label have a privacy policy indicating otherwise. We provide insights into potential sources for discrepancies, including the use of templates and confusion around Apple's definitions and requirements. These results suggest that significant work is still needed to help developers more accurately label their apps. Our system can be incorporated as a first-order check to inform developers when privacy labels are possibly misapplied.

This work reports on a large scale, longitudinal analysis of the adoption dynamics of privacy labels in the iOS App Store, measuring this first-of-its kind ecosystem as it reaches maturity over two and a half years after launching in December 2020. The motivation is to shed light on the factors affecting the shifts in privacy labels and provide insights into how and when an app's label changes. By collecting nearly weekly snapshots of over 1.6 million apps for over a year, we analyze the dynamics of privacy label adoption and the accuracy of reported labels. Our analysis of 74.5% of apps having labels after two years provides important context into this mature ecosystem where labels are becoming the standard. However, we find compelling evidence that labels may not fully capture behavior, as 28.9% of apps indicate no data collection and distributions differ between voluntary versus mandatory adoptions. Once set, labels rarely change but additions reflect more data collection. In addition to our measurement, we also plan to release a new (and growing) data set that can be used by future researchers.

In this paper, we revisit the use of honeypots for detecting reflective amplification attacks. These measurement tools require careful design of both data collection and data analysis including cautious threshold inference. We survey common amplification honeypot platforms as well as the underlying methods to infer attack detection thresholds and to extract knowledge from the data. By systematically exploring the threshold space, we find most honeypot platforms produce comparable results despite their different configurations. Moreover, by applying data from a large-scale honeypot deployment, network telescopes, and a real-world baseline obtained from a leading DDoS mitigation provider, we question the fundamental assumption of honeypot research that convergence of observations can imply their completeness. Conclusively we derive guidance on precise, reproducible honeypot research, and present open challenges.

Targeted advertising is a pervasive practice in the advertising ecosystem, with complex representations of user identity central to targeting. Ad networks are incentivized to tie ephemeral cookies across devices to lasting durable identifiers such as email addresses in order to develop comprehensive cross-device user profiles. Third-party ad networks typically do not have relationships with users and must rely on external parties such as merchant websites for durable identity information, introducing intricate trust relationships. We find attackers can exploit these trust relationships to confuse an ad network into linking an unprivileged attacker's browser to a victim's identity, thus "impersonating" the victim to the ad network.

Single Sign-On (SSO) is both a core and critical component of user authentication and authorization on the modern web, as it is often offered by web and mobile applications alongside credential-based authentication to facilitate the account creation and login process. However, the interplay between local account management and SSO functionality in the backend leads to flaws that enable or magnify account hijacking attacks. These flaws are not baked into the actual SSO protocols, but manifest due to the complexity of supporting separate but intermingling authentication paths. As a result, these types of flaws cannot be detected by the SSO protocol or implementation verification tools proposed in prior work. In this paper we introduce SAAT, a fully automated modular framework that assesses whether relying parties (RPs) that use Facebook as the IdP comply with secure practices and guidelines, and uncovers flaws in account and session management that stem from or are affected by the interplay of SSO and local functionality. We conduct a large-scale exploration of authentication and session practices in Facebook’s RPs, revealing a volatile ecosystem where SSO support can be suddenly dropped and 17.6% of the tested RPs exhibit non-functional SSO implementations. This highlights the need for the continuous and systematic testing of the SSO ecosystem made possible by SAAT. More critically, we find that security measures are often missing and official guidelines are routinely overlooked or misconfigured, with only 0.8% of the RPs fully enabling re-authentication which can prevent compromise from hijacked identity provider (IdP) cookies. Our study also shows that less than 2% of RPs correctly react to SSO revocation and 67% continue to allow account access even 10 days after revocation. Overall, we envision our framework as a tool for enabling and guiding widespread remediation efforts by major SSO identity providers, which were previously infeasible due to the sheer scale and inherent mutability of this ecosystem.

In this paper, we introduce a framework to observe RPKI relying parties (i.e., those that fetch RPKI data from the distributed repository) and present insights into this ecosystem for the first time. Our longitudinal study of data gathered from three RPKI certification authorities (AFRINIC, APNIC, and our own CA) identifies different deployment models of relying parties and (surprisingly) prevalent inconsistent fetching behavior that affects Internet routing robustness. Our results reveal nearly 90% of relying parties are unable to connect to delegated publication points under certain conditions, which leads to erroneous invalidation of IP prefixes and likely widespread loss of network reachability.

When users post on social media, they protect their privacy by choosing an access control setting that is rarely revisited. Changes in users' lives and relationships, as well as social media platforms themselves, can cause mismatches between a post's active privacy setting and the desired setting. The importance of managing this setting combined with the high volume of potential friend-post pairs needing evaluation necessitate a semi-automated approach. We attack this problem through a combination of a user study and the development of automated inference of potentially mismatched privacy settings. A total of 78 Facebook users reevaluated the privacy settings for five of their Facebook posts, also indicating whether a selection of friends should be able to access each post. They also explained their decision. With this user data, we designed a classifier to identify posts with currently incorrect sharing settings. This classifier shows a 317% improvement over a baseline classifier based on friend interaction. We also find that many of the most useful features can be collected without user intervention, and we identify directions for improving the classifier's accuracy.

In this paper, we propose a deep learning framework for user fingerprinting via mobile motion sensors, DeepFP, which can identify and track users based on their behavioral patterns while interacting with the smartphone. Existing machine learning techniques for user identification are classification-oriented and thus are not amenable easily to large-scale, real world deployment. They need to be trained on all the users whom they want to identify. DeepFP exploits metric learning techniques and deep neural networks to address the challenges of current user identification techniques. We leverage feature embedding to directly extract informative features and map input samples to a discriminative lower-dimensional space, where recurrent neural networks are used to model the temporal information of data. DeepFP does not need to re-train to identify new users which makes it feasible to be used in real world scenarios with a huge number of users, without needing a large number of training samples. Experiments on a publicly available mobile sensors dataset and comparison with other embedding methods depict the effectiveness of DeepFP.

It is no secret that users have difficulty choosing and remembering strong passwords, especially when asked to choose different passwords across different accounts. While research has shed light on password weaknesses and reuse, less is known about user motivations for following bad password practices. Understanding these motivations can help us design better interventions that work with the habits of users and not against them.

Users have accumulated years of personal data in cloud storage, creating potential privacy and security risks. This agglomeration includes files retained or shared with others simply out of momentum, rather than intention. We presented 100 online-survey participants with a stratified sample of 10 files currently stored in their own Dropbox or Google Drive accounts. We asked about the origin of each file, whether the participant remembered that file was stored there, and, when applicable, about that file's sharing status. We also recorded participants' preferences moving forward for keeping, deleting, or encrypting those files, as well as adjusting sharing settings. Participants had forgotten that half of the files they saw were in the cloud. Overall, 83% of participants wanted to delete at least one file they saw, while 13% wanted to unshare at least one file. Our combined results suggest directions for retrospective cloud data management.

With the increasing volume of transactions taking place online, mobile fraud has also increased. Mobile applications often authenticate the user only at install time. The user may then remain logged in for hours or weeks. Any unauthorized access may lead to financial, criminal or privacy losses. In this work, we leverage currently available built-in motion sensors in smartphones to learn users' behavioral characteristics while interacting with the mobile device to provide an implicit re-authentication mechanism that enables a frictionless and secure user experience in the application. This approach improves the generality as well as power efficiency of the authentication mechanism compared to using the camera feed which involves (a) specific hardware, (b) higher battery usage and (c) privacy concerns. We present DeepAuth as a generic framework for re-authenticating users in a mobile app. In our approach, we use time and frequency domain features extracted from motion sensors and a long short-term memory (LSTM) model with negative sampling to build a re-authentication framework. The framework is able to re-authenticate a user with 96.70% accuracy in 20 seconds from a set of data collected from 47 volunteers.

Typosquatting is a blackhat practice that relies on human error and low-cost domain registrations to hijack legitimate traffic from well-established websites. The technique is typically used for phishing, driving traffic towards competitors or disseminating indecent or malicious content and as such remains a concern for businesses. We take a fresh new look at this well-studied phenomenon to explore why some URLs are more vulnerable to typing mistakes than others. We explore the relationship between human hand anatomy, keyboard layouts and typing mistakes using various URL datasets. We create an extensive user-centric typographical model and compute a Hardness Quotient (likelihood of mistyping) for each URL using a quantitative measure for finger and hand effort. Furthermore, our model predicts the most likely typos for each URL which can then be defensively registered. Cross-validation against actual URL and DNS datasets suggests that this is a meaningful and effective defense mechanism.

Modern web browsers have accrued an incredibly broad set of features since being invented for hypermedia dissemination in 1990. Many of these features benefit users by enabling new types of web applications. However, some features also bring risk to users' privacy and security, whether through implementation error, unexpected composition, or unintended use. Currently there is no general methodology for weighing these costs and benefits. Restricting access to only the features which are necessary for delivering desired functionality on a given website would allow users to enforce the principle of lease privilege on use of the myriad APIs present in the modern web browser.

Doxing is online abuse where a malicious party harms another by releasing identifying or sensitive information. Motivations for doxing include personal, competitive, and political reasons, and web users of all ages, genders and internet experience have been targeted. Existing research on doxing is primarily qualitative. This work improves our understanding of doxing by being the first to take a quantitative approach. We do so by designing and deploying a tool which can detect dox files and measure the frequency, content, targets, and effects of doxing on popular dox-posting sites.

Billions of people use cloud-based storage for personal files. While many are likely aware of the extent to which they store information in the cloud, it is unclear whether users are fully aware of what they are storing online. We recruited 30 research subjects from Craigslist to investigate how users interact with and understand the privacy issues of cloud storage. We studied this phenomenon through surveys, an interview, and custom software which lets users see and delete their photos stored in the cloud. We found that a majority of users stored private photos in the cloud that they did not intend to upload, and a large portion also chose to permanently delete some of the offending images. We believe our study highlights a mismatch between user expectation and reality. As cloud storage is plentiful and ubiquitous, effective tools for enabling risk self-assessment are necessary to protect users' privacy.

A user's primary email account, in addition to being an easy point of contact in our online world, is increasingly being used as a single point of failure for all web security. Features like unlimited message storage, numerous weak password reset features and economically enticing spoils (in the form of financial accounts or personal photos) all add up to an environment where overthrowing someone's life via their primary email account is increasingly likely and damaging. We describe an attack we call credential based privilege escalation, and a methodology to evaluate this attack's potential for user harm at web scale. In a study of over 9,000 users we find that, unsurprisingly, access to a vast number of online accounts can be gained by breaking into a user's primary email account (even without knowing the email account's password), but even then the monetizable value in a typical account is relatively low. We also describe future directions in understanding both the technical and human aspects of credential based privilege escalation.

While we have a good understanding of how cyber crime is perpetrated and the profits of the attackers, the harm experienced by humans is less well understood, and reducing this harm should be the ultimate goal of any security intervention. This paper presents a strategy for quantifying the harm caused by the cyber crime of typo squatting via the novel technique of intent inference. Intent inference allows us to define a new metric for quantifying harm to users, develop a new methodology for identifying typo squatting domain names, and quantify the harm caused by various typo squatting perpetrators. We find that typo squatting costs the typical user 1.3 seconds per typo squatting event over the alternative of receiving a browser error page, and legitimate sites lose approximately 5% of their mistyped traffic over the alternative of an unregistered typo. Although on average perpetrators increase the time it takes a user to find their intended site, many typo squatters actually improve the latency between a typo and its correction, calling into question the necessity of harsh penalties or legal intervention against this flavor of cyber crime.

E-mail spam has been the focus of a wide variety of measurement studies, at least in part due to the plethora of spam data sources available to the research community. However, there has been little attention paid to the suitability of such data sources for the kinds of analyses they are used for. In spite of the broad range of data available, most studies use a single "spam feed" and there has been little examination of how such feeds may differ in content. In this paper we provide this characterization by comparing the contents of ten distinct contemporaneous feeds of spam-advertised domain names. We document significant variations based on how such feeds are collected and show how these variations can produce differences in findings as a result.

Measurement and analysis of modern malware systems such as botnets relies crucially on execution of specimens in a setting that enables them to communicate with other systems across the Internet. Ethical, legal, and technical constraints however demand containment of resulting network activity in order to prevent the malware from harming others while still ensuring that it exhibits its inherent behavior. Current best practices in this space are sorely lacking: measurement researchers often treat containment superficially, sometimes ignoring it altogether. In this paper we present GQ, a malware execution "farm" that uses explicit containment primitives to enable analysts to develop containment policies naturally, iteratively, and safely. We discuss GQ's architecture and implementation, our methodology for developing containment policies, and our experiences gathered from six years of development and operation of the system.

Spam-based advertising is a business. While it has engendered both widespread antipathy and a multi-billion dollar anti-spam industry, it continues to exist because it fuels a profitable enterprise. We lack, however, a solid understanding of this enterprise's full structure, and thus most anti-Spam interventions focus on only one facet of the overall spam value chain (e.g., spam filtering, URL blacklisting, site takedown).In this paper we present a holistic analysis that quantifies the full set of resources employed to monetize spam email -- including naming, hosting, payment and fulfillment -- using extensive measurements of three months of diverse spam data, broad crawling of naming and hosting infrastructures, and over 100 purchases from spam-advertised sites. We relate these resources to the organizations who administer them and then use this data to characterize the relative prospects for defensive interventions at each link in the spam value chain. In particular, we provide the first strong evidence of payment bottlenecks in the spam value chain, 95% of spam-advertised pharmaceutical, replica and software products are monetized using merchant services from just a handful of banks.

The of spam--the probability that an unsolicited e-mail will ultimately elicit a sale--underlies the entire spam value proposition. However, our understanding of this critical behavior is quite limited, and the literature lacks any quantitative study concerning its true value. In this paper we present a methodology for measuring the conversion rate of spam. Using a parasitic infiltration of an existing botnet's infrastructure, we analyze two spam campaigns: one designed to propagate a malware Trojan, the other marketing on-line pharmaceuticals. For nearly a half billion spam e-mails we identify the number that are successfully delivered, the number that pass through popular anti-spam filters, the number that elicit user visits to the advertised sites, and the number of sales and infections produced.