AI is changing biological and nuclear risks; governance must change accordingly

- This article from the Bulletin of the Atomic Scientists (the group behind the Doomsday Clock) argues that AI is fundamentally changing two existing catastrophic-risk domains: bioweapons and nuclear weapons. - The key idea is "convergence" — AI tools (like large language models or protein-folding AIs) can lower the skill barrier for designing novel pathogens, while AI-enabled drones, surveillance, and autonomous systems erode the stability of nuclear deterrence (e.g., by making second-strike capabilities more vulnerable or speeding up decision-making to dangerous levels). - The Bulletin has long tracked nuclear risk; this piece extends that lens to AI-bio and AI-nuclear intersections, urging that governance frameworks (treaties, export controls, international norms) must evolve as fast as the technology, not lag behind as they did with the early internet or gene editing. - Relevant prior context: the 2023 Executive Order on AI Safety, the Bletchley Park AI Safety Summit, and ongoing debates about "AI biosecurity" (e.g., whether models like GPT-4 can help a non-expert synthesize a pathogen) and "lethal autonomous weapons" (LAWS) talks at the UN.