AI-Nuclear Workstream

The Frontier Model Forum’s (FMF) AI-Nuclear workstream aims to advance the safety and security of leading AI models with respect to nuclear and radiological threats. While frontier AI holds great promise for revolutionizing peaceful nuclear applications in areas like clean energy and medicine, it may be misused by malicious actors to create harm, such as in targeting facilities where radiological or nuclear materials are processed. Understanding and managing these risks are urgent challenges for both the AI and nuclear security communities.

The AI-Nuclear workstream aims to develop shared understandings of threat models, evaluations, and mitigation measures for frontier AI risks related to dual-use nuclear applications. It will also work to identify priority areas for further scientific studies, research methods, and safety practices with respect to frontier AI and nuclear threats. 

The workstream draws on the expertise of the FMF’s member firms, while maintaining strong collaboration with external scientific and policy experts, including leading researchers and practitioners across the nuclear security, non-proliferation, and disarmament fields. By leveraging collective knowledge from academia, industry, and government, the AI-Nuclear workstream remains grounded in the latest empirical and theoretical insights from across sectors and disciplines.

While the FMF is committed to information-sharing and transparency, publications related to AI and the creation of nuclear or radiological threats may introduce significant information and attention hazards. Materials related to this workstream will be released publicly only after careful review and deliberation by domain experts.

Newsroom 

Axios, August 2025. Anthropic, DOE team up to spot dangerous nuclear chats.