The present and future transition of lives and activities into virtual worlds --- worlds in which people interact using avatars --- creates novel privacy challenges and opportunities. Avatars present an opportunity for people to control the way they are represented to other users and the information shared or implied by that representation. Importantly, users with marginalized identities may have a unique set of concerns when choosing what information about themselves (and their identities) to conceal or expose in an avatar. We present a theoretical basis, supported by two empirical studies, to understand how marginalization impacts the ways in which people create avatars and perceive others' avatars: what information do people choose to reveal or conceal, and how do others react to these choices? In Study 1, participants from historically marginalized backgrounds felt more concerned about being devalued based on their identities in virtual worlds, which related to a lower desire to reveal their identities in an avatar, compared to non-marginalized participants. However, in Study 2 participants were often uncomfortable with others changing visible characteristics in an avatar, weighing concerns about others' anonymity with possible threats to their own safety and security online. Our findings demonstrate asymmetries in what information people prefer the self vs. others to reveal in their online representations: participants want privacy for themselves but to feel informed about others. Although avatars allow people to choose what information to reveal about themselves, people from marginalized backgrounds may still face backlash for concealing components of their identities to avoid harm.

We applied techniques from psychology --- typically used to visualize human bias --- to facial analysis systems, providing novel approaches for diagnosing and communicating algorithmic bias. First, we aggregated a diverse corpus of human facial images (N=1492) with self-identified gender and race. We tested four automated gender recognition (AGR) systems and found that some exhibited intersectional gender-by-race biases. Employing a technique developed by psychologists --- face averaging --- we created composite images to visualize these systems' outputs. For example, we visualized what an "average woman" looks like, according to a system's output. Second, we conducted two online experiments wherein participants judged the bias of hypothetical AGR systems. The first experiment involved participants (N=228) from a convenience sample. When depicting the same results in different formats, facial visualizations communicated bias to the same magnitude as statistics. In the second experiment with only Black participants (N=223), facial visualizations communicated bias significantly more than statistics, suggesting that face averages are meaningful for communicating algorithmic bias.

Online platforms employ manual human moderation to distinguish human-created social media profiles from deepfake-generated ones. Biased misclassification of real profiles as artificial can harm general users as well as specific identity groups; however, no work has yet systematically investigated such mistakes and biases. We conducted a user study (n=695) that investigates how 1) the identity of the profile, 2) whether the moderator shares that identity, and 3) components of a profile shown affect the perceived artificiality of the profile. We find statistically significant biases in people’s moderation of LinkedIn profiles based on all three factors. Further, upon examining how moderators make decisions, we find they rely on mental models of AI and attackers, as well as typicality expectations (how they think the world works). The latter includes reliance on race/gender stereotypes. Based on our findings, we synthesize recommendations for the design of moderation interfaces, moderation teams, and security training.

Research in computer security has increasingly considered the needs of marginalized and vulnerable groups in technology. Through this work, we hope to translate this research movement into practice and, ultimately, cause designers-in-training (and, eventually, designers) to consider a more inclusive range of stakeholders. Thus, we created an educational intervention to center marginalized and vulnerable populations in the context of threat modeling. We find that computer security students are more likely to consider unique threats and vulnerabilities facing marginalized and vulnerable populations after being exposed to an intervention prompting them to think about populations that might often be overlooked. We suggest practical methods to teach designers-in-training inclusive methods in computer security and discuss other possible adoptions of this practice across the field. This work is part of an important shift toward inclusive security that centers marginalized and vulnerable populations both in research and in practice.

Parents posting photos and other information about children on social media is increasingly common and a recent source of controversy. We investigated characteristics that predict parental sharing behavior by collecting information from 493 parents of young children in the United States on self-reported demographics, social media activity, parenting styles, children's social media engagement, and parental sharing attitudes and behaviors. Our findings indicate that most social media active parents share photos of their children online and feel comfortable doing so without their child's permission. The strongest predictor of parental sharing frequency was general social media posting frequency, suggesting that participants do not strongly differentiate between "regular" photo-sharing activities and parental sharing. Predictors of parental sharing frequency include greater social media engagement, larger social networks with norms encouraging parental sharing, more permissive and confident parenting styles, and greater social media engagement by their children. Contrasting previous research that often highlights benefits of parental sharing, our findings point to a number of risky online behaviors associated with parental sharing not previously uncovered. Implications for children's privacy and early social media exposure are discussed, including future directions for influencing parental sharing attitudes and behaviors.

We investigate how people’s ‘humor style’ relates to their online photo-sharing behaviors and reactions to ‘privacy primes’. In an online experiment, we queried 437 participants about their humor style, likelihood to share photo-memes, and history of sharing others’ photos. In two treatment conditions, participants were either primed to imagine themselves as the photo-subjects or to consider the photo-subjects’ privacy before sharing memes. We found that participants who frequently use aggressive and self-deprecating humor were more likely to violate others’ privacy by sharing photos. We also replicated the interventions’ paradoxical effects – increasing sharing likelihood – as reported in earlier work and identified the subgroups that demonstrated this behavior through interaction analyses. When primed to consider the subjects’ privacy, only humor deniers (participants who use humor infrequently) demonstrated increased sharing. In contrast, when imagining themselves as the photo-subjects, humor deniers, unlike other participants, did not increase the sharing of photos.

We investigate the effects of perspective taking, privacy cues, and portrayal of photo subjects (i.e., photo valence) on decisions to share photos of people via social media. In an online experiment we queried 379 participants about 98 photos (that were previously rated for photo valence) in three conditions: (1) Baseline: participants judged their likelihood of sharing each photo; (2) Perspective-taking: participants judged their likelihood of sharing each photo when cued to imagine they are the person in the photo; and (3) Privacy: participants judged their likelihood to share after being cued to consider the privacy of the person in the photo. While participants across conditions indicated a lower likelihood of sharing photos that portrayed people negatively, they - surprisingly - reported a higher likelihood of sharing photos when primed to consider the privacy of the person in the photo. Frequent photo sharers on real-world social media platforms and people without strong personal privacy preferences were especially likely to want to share photos in the experiment, regardless of how the photo portrayed the subject. A follow-up study with 100 participants explaining their responses revealed that the Privacy condition led to a lack of concern with others' privacy. These findings suggest that developing interventions for reducing photo sharing and protecting the privacy of others is a multivariate problem in which seemingly obvious solutions can sometimes go awry.