Redeploying Fable 5
Anthropic News · 2026-06-30
Anthropic announces the redeployment of Claude Fable 5 after US government export controls imposed on June 12, 2026, were lifted on June 30, and proposes a new industry-wide framework for scoring AI jailbreak severity alongside expanded government collaboration commitments.
Appears in
Extraction
Topics: ai-model-safetyexport-controlsjailbreak-severity-frameworkcybersecurity-safeguardsgovernment-ai-collaboration
Claims
- The US government lifted export controls on Claude Fable 5 and Mythos 5 on June 30, 2026, after imposing them on June 12 following a report by Amazon researchers of a safeguard bypass.
- The reported jailbreak technique did not expose unique Mythos-level cyber capabilities; the same vulnerability identification and exploit demonstration were replicable by many less capable models including Claude Haiku 4.5, GPT-5.5, and Kimi K2.7.
- Anthropic deployed an improved safety classifier targeting the reported bypass that blocks it in over 99% of cases, at the cost of increased false positives on benign coding and debugging tasks.
- Anthropic is partnering with Amazon, Microsoft, Google, and other Glasswing partners to draft a consensus industry framework for assessing AI jailbreak severity across four dimensions: capability gain, breadth of capability gain, ease of weaponization, and discoverability.
- Anthropic has committed to pre-release government access and independent evaluation for frontier models with national security relevance, rapid information sharing on significant jailbreaks, and dedicated joint research resources with US government partners.
Key quotes
There's currently no consensus in the AI industry on how to describe, in objective terms, the severity of an AI jailbreak. This adds a great deal of uncertainty whenever a new jailbreak technique is discovered: developers have no agreed-upon standard for which findings to focus on most urgently, and governments have no agreed-upon standard for when to act.
Our testing confirmed that many less capable models—including Claude Opus 4.8, GPT-5.5, and Kimi K2.7—could identify the same vulnerabilities as Fable 5 did in the report. When it came to the demonstration of how to exploit the single vulnerability, every model we tested could produce the same demonstration as Fable 5.
We seek to ensure that we and our safety partners will be the first to find major jailbreaks and fix them before malicious actors can use them for harm.