Is Global AI Safety Converging? AI Governance in the UK, EU & Korea | Panel (Alignment Workshop)
A panel on whether global AI safety governance is converging, with the UK AI Security Institute, the EU AI Office, and Korea's AI Safety Institute. Moderated by Isabella of the Safe AI Forum, the panel brings together Myungjoo Kim (Korea AISI), Lisa (EU AI Office), Stephen Casper (Harvard, formerly UK AISI), and Cozmin Ududec (UK AI Security Institute) to compare how jurisdictions approach frontier AI. They map each institution's mandate and risk taxonomy, covering loss of control, CBRN, cyber offense, and harmful manipulation, and debate whether mandatory evaluation should trigger on compute thresholds or on a layered mix of capability, autonomy, and deployment context. The conversation turns to the value of regulatory experimentation, the position of the US, concentration-of-power risk as frontier model weights come from a few actors, and the growing role of third-party evaluators. Chapters 0:00 Introductions and the central question 1:11 Korea AISI: mandate, sovereign AI, risk map (MJ Kim) 4:04 EU AI Office: the AI Act and Code of Practice (Lisa) 6:55 The US question and academia's role (Casper) 8:42 UK AISI: tracking frontier capabilities (Ududec) 10:53 What does each institution focus on? 19:29 Where are jurisdictions converging? 25:58 Convergence in practice: a Korea-Singapore joint evaluation 29:01 Too much bandwagoning? The case for experimentation 32:49 How AI governance could look different in 5 years 35:40 What divergence most needs resolving? 39:45 The position of the US 41:18 Compute thresholds vs. layered risk triggers 43:24 Audience Q: what would governance failure look like? 49:52 Audience Q: concentration of power and frontier weights 54:19 Audience Q: encouraging regulatory experimentation 57:00 Closing: hopes for global coordination More AI safety research: https://far.ai Alignment Workshop playlist: https://youtube.com/playlist?list=PLBY5kyt_LfFg&si=kalx9Z6CdGRc2h21 FAR.AI is a research nonprofit working to ensure the safe development of advanced AI. We host the Alignment Workshop series and publish frontier alignment research.