As AI evolves from models that answer questions into agents that can plan, call tools, and carry out tasks on their own, the boundaries of AI security are shifting with it. Traditional identity management, which checks credentials at the point of entry, can no longer keep track of what an agent does once it is running. Models, data, and secrets entrusted to cloud and accelerated computing environments need protection that can be verified. LLMs hallucinate and behave unpredictably, so AI safety can’t simply inherit the mindset of conventional cybersecurity. And the dangerous behaviors a model might exhibit need to be systematically evaluated before it is ever deployed.

This forum takes on these challenges one by one, from infrastructure and model behavior to evaluation methods. Experts from industry and academia will share how delegated authorization, distinct agent identities, and dynamic, short-lived credentials can be used to make real-time decisions on every single tool call. They will also explain how Trusted Execution Environments (TEEs) and remote attestation can verify a computing environment before any sensitive assets are released to it. Speakers will revisit the very nature of AI safety through the lens of LLM hallucination and break down the methods and limitations of capability, propensity, and control evaluations. Meeting these challenges will take more people working on AI safety. The forum will feature a video message from Ambassador-at-Large Audrey Tang on the role Taiwan can play in AI safety, and NTU AI Safety will share resources and pathways for getting involved in the field.

Everyone interested in AI safety is warmly invited to join us at the "Advanced AI Safety Forum: Research and Technology Frontiers (2026)", co-hosted by the NTU IoX Center, the Taiwan AI Center of Excellence (Taiwan AICoE), the  Taiwan Academic Cybersecurity Center (TACC), and NTU AI Safety. The forum will take place on December 6, 2026, at Tsai Lecture Hall(霖澤館), College of Law, National Taiwan University.

Date: Sunday, December 6, 2026
Time: 8:30 AM – 4:30 PM
Venue: Tsai Lecture Hall(霖澤館), College of Law, National Taiwan University
Format: In-person only; advance registration required. The organizers reserve the right to accept or decline registrations.
Website: https://sites.google.com/g.ntu.edu.tw/2026-advanced-ai-safety-forum
Registration Closed: Friday, November 27, 2026
Registration: https://forms.gle/ZNXqMqHBNbxze6hf8