OpenAI and Frontier Model Forum publish jailbreak rating framework for cyber harm

Why it matters

OpenAI joined Frontier Model Forum members on a jailbreak rating framework meant to compare how severe a cyber-capable bypass is and how readily it can be reused. Policy lead Sasha Baker cast it as shared evidence for the AI and security communities rather than a single-lab score. The FMF page is the primary write-up; OpenAI’s evening post drove the noise spike. Watch whether other frontier labs publish scored examples against the same rubric next week.