The Frontier Model Forum is an industry-supported non-profit rather than a commercial product. It was launched by Anthropic, Google, Microsoft and OpenAI to make sure the most advanced AI systems remain safe and secure, and it draws on the technical and operational expertise of those member companies. Its stated focus is on significant risks to public safety and national security rather than on near-term commercial deployment.
The Forum operates through three core mandates. First, it identifies best practices and supports standards development for frontier AI safety and security, building shared understanding of threat models, evaluations, thresholds and mitigations. Second, it advances the science and supports independent research, including standardised capability and safety evaluations. Third, it facilitates information sharing among government, academia, civil society and industry.
Practically, this work is organised into workstreams covering AI-Bio, AI-Cyber, AI-Nuclear and AI Security, alongside a 'Frontier Frameworks' track, and is funded in part through an AI Safety Fund. Published outputs are technical and policy-facing: for example, a technical report on Frontier Capability Assessments, which examines procedures used to detect capabilities that could raise public-safety or national-security risks, along with issue briefs and public comments on AI policy. Membership is intended for organisations developing frontier models, and the site asks interested parties to get in touch.
