Abstract In this article we report on a post-mortem analysis of the X-XSS-Protection header. Our analysis used data from header usage in the wild at the time of its death as judged by the Google Chrome deprecation notice. We processed roughly 2.6 billion HTTP responses and classified them not just by validity or invalidity but by the reasons for each. We created a set of regex-driven attributions (RDAs) to support robust classification. We found great security posture with minimum levels of disablement and a strong preference for additional protections. Unfortunately, we also discovered that misconfigurations often resulted in dangerous situations with potentially less than the expressly desired level of security. Our results have implications for the potential reconsideration of what a browser should consider a fail-safe design, the opportunity to design new mechanisms with security configuration gradients, and motivation that solving just a few errors can make an enormous impact at internet scale. Finally, this article motivates the security community to consider conducting additional post-mortem analysis in a variety of contexts besides the deprecated HTTP header presented here.

Many proposed mechanisms for helping people understand the security or privacy of their information involve notifying users at various times regarding how that information is used or shared. However, there is a limit as to how many notifications users could attend to, and increasing the frequency and intrusiveness of the notifications will at some point reduce their effectiveness. We do not yet have good guidelines regarding such tradeoffs for different styles of notifications. In this article, we compare users’ reactions and perceived usefulness of notifications of information access on a smart phone, testing the boundaries of acceptability for four different modes of notifications. Our results indicate that the notifications did increase participants’ awareness of information access, but that even for relatively passive notifications, people became annoyed at fairly low frequencies, and phone use was reduced as frequencies increased.

Mobile devices and third-party applications are used by over 4.5 billion people worldwide. Third-party applications often request or even require authorized access to personal information through mobile device components. Application developers explain the need for access in their privacy policies, yet many users are concerned about the privacy implications of allowing access to their personal information. This article explores how user perceptions of privacy affect user sentiment by analyzing over five million user-submitted text reviews and star ratings collected over a four-year period. The authors use supervised machine learning to classify privacy and non-privacy-related reviews. The authors then use natural language processing sentiment analysis to compare differences between the groups. Additionally, the article explores various aspects of both privacy and non-privacy-related reviews using self-reported measurements such as star rating and helpfulness tags.

Managing the privacy of online information can be a complex task often involving the configuration of a variety of settings. For example, Facebook users determine which audiences have access to their profile information and posts, how friends can interact with them through tagging, and how others can search for them—and many more privacy tasks. In most cases, the default privacy settings are permissive and appear to be designed to promote information sharing rather than privacy. Managing privacy online can be complex and often users do not change defaults or use granular privacy settings. In this article, we investigate whether default privacy settings on social network sites could be more customized to the preferences of users. We survey users' privacy attitudes and sharing preferences for common SNS profile items. From these data, we explore using audience characterizations of profile items to quantify fit scores that indicate how well default privacy settings represent user privacy preferences. We then explore the fit of various schemes, including examining whether privacy attitude segmentation can be used to improve default settings. Our results suggest that using audience characterizations from community data to create default privacy settings can better match users' desired privacy settings.

Users are sharing and consuming enormous amounts of information through online social network interaction every day. Yet, many users struggle to control what they share to their overlapping social spheres. Google+ introduces circles, a mechanism that enables users to group friends and use these groups to control their social network feeds and posts. We present the results of a qualitative interview study on the sharing perceptions and behavior of 27 Google+ users. These results indicate that many users have a clear understanding of circles, using them to target information to those most interested in it. Yet, despite these positive perceptions, there is only moderate use of circles to control information flow. We explore reasons and risks associated with these behaviors and provide insight on the impact and open questions of this privacy mechanism.

Photo tagging is a popular feature of many social network sites that allows users to annotate uploaded images with those who are in them, explicitly linking the photo to each person's profile. In this paper, we examine privacy concerns and mechanisms surrounding these tagged images. Using a focus group, we explored the needs and concerns of users, resulting in a set of design considerations for tagged photo privacy. We then designed a privacy enhancing mechanism based on our findings, and validated it using a mixed methods approach. Our results identify the social tensions that tagging generates, and the needs of privacy tools to address the social implications of photo privacy management.

Social navigation is a promising approach to help users make better privacy and security decisions using community knowledge and expertise. Social navigation has recently been applied to several privacy and security systems such as peer-to-peer file sharing, cookie management, and firewalls. However, little empirical evaluation of social navigation cues has been performed in security or privacy systems to understand the real impact such knowledge has on user behavior and the resulting policies. In this paper, we explore the application of social navigation to access control policy configuration using an empirical between subjects study. Our results indicate that community information does impact user behavior, but only when the visual representation of the cue is sufficiently strong.

Online social network sites, such as MySpace, Facebook and others have grown rapidly, with hundreds of millions of active users. A new feature on many sites is social applications -- applications and services written by third party developers that provide additional functionality linked to a user's profile. However, current application platforms put users at risk by permitting the disclosure of large amounts of personal information to these applications and their developers. This paper formally abstracts and defines the current access control model applied to these applications, and builds on it to create a more secure framework. We do so in the interest of preserving as much of the current architecture as possible, while seeking to provide a practical balance between security and privacy needs of the users, and the needs of the applications to access users' information. We present a user study of our interface design for setting a user-to-application policy. Our results indicate that the model and interface work for users who are more concerned with their privacy, but we still need to explore alternate means of creating policies for those who are less concerned.