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

The Bulletin of the Atomic Scientists | Stephen Herzog, Allison Berke, Yanliang Pan, William C. Potter, Douglas B. Shaw

In April, over 100 experts convened at California’s Asilomar Conference Grounds to address how artificial intelligence (AI) may affect nuclear and biological weapons, launching a new Asilomar Process. This initiative responds to rapidly advancing AI models, which pose risks like assisting novices in developing chemical or biological weapons and compressing nuclear decision-making times, increasing launch pressure from false information.

A fundamental governance gap exists where private AI developers identify dangerous capabilities first but lack security expertise, while nuclear and biological security experts need AI insights. The conference adopted seven principles to bridge this divide, emphasizing human control over nuclear weapons, anticipatory risk governance by AI developers, and globally inclusive collaboration aligned with the NPT and BWC. The Asilomar Process seeks to establish collaborative thresholds for emerging AI capabilities, informing national and international policy to prevent AI-enabled nuclear or biological catastrophes before critical decisions are made.

To read in details click here