All major web browsers support extensions to provide additional functionalities and enhance users’ browsing experience while the extensions can access and collect users’ data during their web browsing. Although the web extensions inform users of their data practices via multiple forms of notices, prior work has overlooked the critical gap between the actual data practices and the published privacy notices of browser extensions. To fill this gap, we propose ExtPrivA that automatically detects the inconsistencies between browser extensions’ data collection and their privacy disclosures. From the privacy policies and Dashboard disclosures, ExtPrivA extracts privacy statements to have a clear interpretation of the privacy practices of an extension. It emulates user interactions to trigger the extension’s functionalities and analyzes the initiators of network requests to accurately extract the users’ data transferred by the extension from the browser to external servers. Our end-to-end evaluation has shown ExtPrivA to detect inconsistencies between the privacy disclosures and data-collection behavior with an 85% precision. In a large-scale study of 47.2k extensions on the Chrome Web Store, we found 820 extensions with 1,290 flows that are inconsistent with their privacy statements. Even worse, we have found 525 pairs of contradictory privacy statements in the Dashboard disclosures and privacy policies of 360 extensions. These discrepancies between the privacy disclosures and the actual data-collection behavior are deemed as serious violations of the Store’s policies. Our findings highlight the critical issues in the privacy disclosures of browser extensions that potentially mislead, and even pose high privacy risks to, end-users.

Online trackers, such as advertising and analytics services, have provided users with choices to opt out of their tracking and data collection to mitigate the users' concerns about increased privacy risks. While opt-out choices of online services for the cookies placed on their own websites have been examined before, the choices provided by trackers for their third-party tracking services on publisher websites have been largely overlooked. There is no guarantee that a tracker's opt-out options would faithfully follow the statements in its privacy policy. To address this concern, we develop an automated framework, called OptOutCheck, that analyzes (in)consistencies between trackers' data practices and the opt-out choice statements in their privacy policies. We create sentence-level classifiers, which achieve ≥ 84.6% precision on previously-unseen statements, to extract the opt-out policies that state neither tracking nor data collection for opted-out users from trackers' privacy-policy documents. tOptOutCheck analyzes both tracker and publisher websites to detect opt-out buttons, perform the opt-out, and extract the data flows to the tracker servers after the user opts out. Finally, we formalize the opt-out policies and data flows to derive logical conditions to detect the inconsistencies. In a large-scale study of 2.9k popular trackers, OptOutCheck detected opt-out choices on 165 trackers and found 11 trackers who exhibited data practices inconsistent with their stated opt-out policies. Since inconsistencies are violations of the trackers' privacy policies and demonstrate data collection without user consent, they are likely to lose users' trust in the online trackers and trigger the necessity of an automatic auditing process.

As social robots become increasingly prevalent in day-to-day environments, they will participate in conversations and appropriately manage the information shared with them. However, little is known about how robots might appropriately discern the sensitivity of information, which has major implications for human-robot trust. As a first step to address a part of this issue, we designed a privacy controller, Confidant, for conversational social robots, capable of using contextual metadata (e.g., sentiment, relationships, topic) from conversations to model privacy boundaries. Afterwards, we conducted two crowdsourced user studies. The first study ($n=174$) focused on whether a variety of human-human interaction scenarios were perceived as either private/sensitive or non-private/non-sensitive. The findings from our first study were used to generate association rules. Our second study ($n=95$) evaluated the effectiveness and accuracy of the privacy controller in human-robot interaction scenarios by comparing a robot that used our privacy controller against a baseline robot with no privacy controls. Our results demonstrate that the robot with the privacy controller outperforms the robot without the privacy controller in privacy-awareness, trustworthiness, and social-awareness. We conclude that the integration of privacy controllers in authentic human-robot conversations can allow for more trustworthy robots. This initial privacy controller will serve as a foundation for more complex solutions.

Abstract Advances in deep learning have made face recognition technologies pervasive. While useful to social media platforms and users, this technology carries significant privacy threats. Coupled with the abundant information they have about users, service providers can associate users with social interactions, visited places, activities, and preferences–some of which the user may not want to share. Additionally, facial recognition models used by various agencies are trained by data scraped from social media platforms. Existing approaches to mitigate associated privacy risks result in an imbalanced trade-off between privacy and utility. In this paper, we address this trade-off by proposing Face-Off, a privacy-preserving framework that introduces strategic perturbations to images of the user’s face to prevent it from being correctly recognized. To realize Face-Off, we overcome a set of challenges related to the black-box nature of commercial face recognition services, and the scarcity of literature for adversarial attacks on metric networks. We implement and evaluate Face-Off to find that it deceives three commercial face recognition services from Microsoft, Amazon, and Face++. Our user study with 423 participants further shows that the perturbations come at an acceptable cost for the users.