Invited Talk (AdvML): Motahhare Eslami — The Future of AI Safety: User-Engaged Auditing and Human-AI Red Teams

Carnegie Mellon University

Abstract: As AI systems become more general-purpose, interactive, and embedded in high-stakes settings, the challenge of auditing them is changing. In this talk, I will trace an evolution in AI safety research from traditional AI auditing to user-engaged auditing and, ultimately, human–AI collaborative red teaming. I will first discuss empirical audits of search engines, rating platforms, and online housing systems that exposed systematic biases, misleading algorithmic behaviors, and discriminatory outcomes. Building on lessons from this work, I will then introduce user-engaged AI auditing, which treats users not simply as people affected by AI failures, but as valuable contributors who can identify, investigate, and contextualize them. I will highlight studies of grassroots auditing practices as well as tools such as MIRAGE, Vipera, and WeAudit that scaffold broader participation in generative AI evaluation. Finally, I will present recent work on Human-Agent Audit Collaboration (HAAC), examining how AI agents can expand the scale and breadth of adversarial testing while also shaping what human auditors explore, notice, and judge as harmful. Together, these efforts point toward a future of AI safety that combines the scale of AI with the contextual judgment, lived experience, and diverse perspectives of human auditors.

Bio: Motahhare Eslami is an Associate Professor at Carnegie Mellon University's School of Computer Science, Human-Computer Interaction Institute (HCII), and Software and Societal Systems Department (S3D). Motahhare’s research is broadly focused on Human-AI Interaction, AI auditing, evaluation, and safety. Motahhare is named one of the 100 Brilliant Women in AI Ethics, and her work has been recognized with multiple awards at top-tier ACM and AAAI conferences. She has also received several awards from industry including the Google Research Scholar Award, the Microsoft AI & Society Fellowship, and the Google Academic Research Award for Society-Centered AI. Her work has been featured in major media outlets like Time, The Washington Post, and WIRED, and is supported by NSF, CMU Block Center, Amazon, Google, Meta, Microsoft, Cisco, BNY and eBay.