FMF jailbreak rating framework for cyber threats
Confirmed
The takeaway
On 5 Nov 2026 the Frontier Model Forum released version 1.0 of a “preliminary” framework that rates successful cyber jailbreaks by comparative risk, scoring capability gain, capability scope, ease of deployment and ease of discovery.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 7 November 2026
Your AI and this story
- GPT-6 Astra189 days after its cutoff
- Claude Opus 5.5128 days after its cutoff
- Gemini 4 Argon219 days after its cutoff
- Grok 4.7158 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 128 days before it.
Key facts
- Four scoring categories combined into tiers: capability gain, capability scope, ease of deployment, ease of discovery
- Contributors: Amazon Web Services, Anthropic, Google DeepMind, Microsoft, OpenAI (supporting statements)
- OpenAI’s Sasha Baker: ‘an important step toward a common methodology for evaluating and comparing the severity of cyber jailbreaks’ (OpenAI X post, ~277k views)
- Comments to comments@frontiermodelforum.org until Dec 21, 2026; the FMF will test assessor consistency
What happened
The FMF published the framework on 5 Nov, the same day OpenAI launched GPT-6 Cyber. It is meant to let labs triage reported jailbreaks consistently.
Why it matters
The labs now have a first shared scale for how bad a safeguard bypass is in the cyber domain, which bug bounties and incident reports can use.
Sources
2 sources from 2 sites. Numbers match the chips in the text.
2 sources: 2 primary
Primary
- FMF: Introducing the FMF jailbreak rating framework for frontier cyber threatsfrontiermodelforum.org, official
- OpenAI on Xx.com, official
Changes
- Filed (lab X list)