Frontier Model Forum Launches Jailbreak Risk Framework

This is less a new defense than a shared triage system—but that standardization could make frontier-model security work substantially more actionable.

• –Gives developers a common vocabulary for comparing jailbreak findings across models and organizations
• –Focuses limited mitigation resources on bypasses that unlock broad, usable cyber capabilities
• –Separates a jailbreak’s relative risk from the total risk of the surrounding AI system
• –Establishes a practical baseline for security researchers, while remaining open to public feedback through December 21, 2026