BBL Speaker Series: Julio Poveda: Insights from Auditing AI Chatbots for Survivors of Domestic Violence

Talk Title: Insights from Auditing AI Chatbots for Survivors of Domestic Violence

Speaker: Julio Poveda, PhD Candidate, Computer Science, University of Maryland.

Location: IRB 4105 and Zoom

Abstract: Domestic violence (DV) is a public health issue that affects people of all genders and socioeconomic statuses worldwide. As DV service providers explore scalable ways to support survivors, AI chatbots are increasingly being deployed for DV support. Yet questions remain about their safety, effectiveness, and governance. In this talk, I will present our work that sought to systematically identify and audit DV chatbots. We found 36 functional DV chatbots developed across 22 countries, offering services ranging from educating on what a healthy relationship is and guiding safety planning. At the same time, we identified various issues around transparency, data gathering and processing practices, and features that can put survivors’ safety at further risk. During the talk I will also share our recommendations for advocacy groups considering adopting these AI chatbots, as well as best practices for developers.

Bio:  Julio Poveda is a Ph.D. candidate in Computer Science at the University of Maryland. He does cybersecurity and privacy research, specifically to improve the digital safety of at-risk populations, including older adults and survivors of domestic violence. Additionally, he volunteers at the Clinic to End Tech Abuse, where he supports survivors of technology-facilitated abuse.