Progress Update: FMF Information-Sharing of Frontier AI Threats and Vulnerabilities
The Frontier Model Forum announced accomplishments in its voluntary information-sharing agreement launched in March 2025, which allows member organizations to exchange data about frontier AI vulnerabilities, threats, and concerning capabilities.
The FMF established protocols for sharing three categories of sensitive information: vulnerabilities like jailbreaks and model safeguard bypasses; threats including unauthorized access attempts and cyber-threat indicators; and capabilities that could enable large-scale harm, such as CBRN development or autonomous systems.
Member firms have successfully exchanged this information, using shared data to strengthen their AI safety and security measures. The organization reports that "legal and technical infrastructure has supported the secure, confidential exchange of threat related information while protecting intellectual property and ensuring antitrust compliance."
The information-sharing mechanism has also supported development of industry best practices and AI safety research, particularly regarding dual-use prompts that could pose cyber and biological risks.
The FMF intends to expand participation beyond current members to other frontier AI companies through pilot programs. The organization plans to establish additional secure channels for threat information sharing across industry and government stakeholders, develop shareable resources addressing information hazards, and create interoperable frameworks as related sharing pathways emerge.
The forum remains dedicated to its founding mandate to "establish trusted, secure mechanisms for sharing information among companies, governments and relevant stakeholders regarding AI safety and risks."