Anthropic details Fable 5’s four-tier cyber classifiers and proposes a Cyber Jailbreak Severity (CJS) scale
Confirmed
The takeaway
On July 2, 2026 Anthropic explained how Claude Fable 5’s cyber classifiers sort requests into four tiers: prohibited, high-risk dual use, low-risk dual use and benign.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 2 of 5
- Last verified
- 1 October 2026
Your AI and this story
- GPT-6.2 Astra32 days after its cutoff
- Claude Opus 5.52 days after its cutoff
- Gemini 4 Hydrogen93 days after its cutoff
- Grok 4.732 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 2 days before it.
Key facts
- Tier 1, prohibited (ransomware, malware development, data exfiltration): blocked
- Tier 2, high-risk dual use (pen-testing, exploit development): blocked pending better access controls/authorization context
- Tier 3, low-risk dual use (e.g. vulnerability identification): sometimes blocked as a safety margin
- Tier 4, benign (patching, incident response): allowed with minimal false positives
- CJS-0 to CJS-4 severity scale over four axes: capability gain, breadth of gain, ease of weaponization, discoverability
What happened
Published about three weeks after Fable 5’s launch, while users complained that cyber classifiers were over-blocking. The post describes the classifier policy and offers a shared vocabulary for rating jailbreaks.
Why it matters
It shows how a lab gates Mythos-class cyber capability in practice. The CJS scale is offered as a cross-industry standard for reporting jailbreak severity, which no one had before.
Unverified: whether other labs or bug-bounty programs adopted CJS.
Sources
1 source from 1 site. Numbers match the chips in the text.
1 source: 1 primary
Primary
- Anthropic: More details on Fable 5’s cyber safeguards and our jailbreak frameworkanthropic.com, official
Changes
- Filed