Brief Description: U.S. consumers reported $12.5 billion in fraud losses in 2025, a 25% jump, and an undercount. This talk looks underneath that number at the machinery scammers use: a small, stable set of psychological techniques such as authority, urgency, trust, fear of loss, and phantom riches, deployed inside recurring scam scenarios that are rebranded and re-channeled continuously. Drawing on an analysis of over 600,000 real scam reports and recent AI research, Dr. Xiao traces a scam conversation turn by turn, shows how scam playbooks change and how fast detection rules go stale, and lays out three lines of defense built on AI: prevention, surveillance, and intervention, plus where this is heading next.

Bio: Xusheng Xiao is an Associate Professor in the School of Computing and Augmented Intelligence at Arizona State University, where he leads the Reliable, Intelligent, Secure, and Efficient (RISE) Software and System Lab. His research sits at the intersection of cybersecurity, software engineering, and artificial intelligence, with a current focus on LLM-powered fraud and deception analysis. His group's PsyScam benchmark (EMNLP 2025) grounds the psychological techniques scammers use in real-world scam reports, and its PreScam benchmark (COLM 2026) tests whether language models can predict how an ongoing scam conversation will progress, early enough for a defender to intervene, while related work characterizes how scam scenarios evolve across time and geography. His earlier research on cyber threat detection and investigation was integrated into a commercial security product at NEC that won the Grand Prix Award at CEATEC 2016, and his industry collaborations have produced six U.S. patents. He is a recipient of the NSF CAREER Award, NSF CRII Award, and a Samsung Global Research Outreach Award, with research supported by NSF, DOE-ARPA-E, Samsung, and the OpenAI Cybersecurity Grant Program. He received his Ph.D. from North Carolina State University in 2014.