Abstract: The rapid rise of generative AI has transformed how young people interact with technology, yet its safety mechanisms remain largely one-size-fits-all, failing to account for developmental needs, social contexts, and the nuanced ways youth experience online risks. These blind spots expose young users to a spectrum of harms, from emotional overdependence on AI companions to misinformation during identity exploration, while leaving families and educators without effective means of oversight. Addressing these challenges requires moving beyond traditional moderation to build human-centered AI safety systems that adapt to users’ unique abilities and social environments.
In this talk, I will present my research on advancing youth-centered AI safety through empirical studies, risk taxonomies, and safeguard system development. I will begin by introducing the Youth AI Risk (YAIR) taxonomy, the first empirically grounded framework that captures 84 distinct risk manifestations from real youth AI interactions. Building on this foundation, I will describe the YouthSafe benchmark and guardrail model, a context aware AI safeguard capable of detecting subtle and developmental risks for youth missed by current moderation tools. I will then introduce YouthSafeAgent, a collaborative safety platform that bridges the gap between youth autonomy and parental awareness by enabling shared safety goals and trust based communication. I will conclude with a brief discussion of how human-centered safety design generalizes to other at-risk users and technologies, from web3 security to accessible AI for people with visual impairments, and outline future directions aimed at advancing next-generation AI safety mechanisms and techniques that can safeguard and empower a wide range of users.
Bio: Yaman Yu is a PhD Candidate in Information Science at the University of Illinois Urbana-Champaign. Her research integrates Human Computer Interaction (HCI) and Cybersecurity to build human-centered AI safety, security, and privacy solutions for at-risk populations. Yaman is a recipient of the 2025 Google PhD Fellowship, was named an MIT EECS Rising Star, and received a CHI Best Paper Honorable Mention award. Her work, published at top tier conferences such as CHI, IEEE S&P, and CCS, has been featured in outlets like the New York Times, Business Insider and MIT Technology Review.