Frontier Model Forum releases FMF jailbreak rating framework 1.0

The Frontier Model Forum has released Version 1.0 of its Jailbreak Rating Framework for Frontier Cyber Threats, a preliminary tool designed to help organizations consistently rate and prioritize successful AI cyber jailbreaks. The framework combines scores across capability gain, capability scope, ease of deployment, and ease of discovery. Developed with contributions from companies including Amazon Web Services, Anthropic, Google DeepMind, Microsoft, and OpenAI, the framework is open for public comment through December 21, 2026. Open article →