Anthropic — Claude Fable 5 Returns With Industry Jailbreak Framework After Export Controls Lifted
AI relevance: The redeployment of a frontier model following government export controls over cybersecurity capability concerns introduces the first cross-industry framework for classifying jailbreak severity — directly affecting how AI ops teams evaluate model safety in production.
- On June 12, the US government applied export controls to Claude Fable 5 and Mythos 5 after Amazon researchers found a method to bypass Fable 5's safeguards and elicit vulnerability-finding behavior, including exploitation code for at least one CVE.
- Anthropic suspended all access to both models immediately because it had no reliable way to verify user nationality in real-time. Access has now been restored globally as of July 1 after controls were lifted on June 30.
- Anthropic's internal testing confirmed that less capable models — including Claude Opus 4.8, GPT-5.5, and Kimi K2.7 — could identify the same vulnerabilities. Every model tested, including Claude Haiku 4.5, could reproduce the exploitation demonstration.
- The reported technique did not expose unique Mythos-level cyber capabilities. Anthropic characterizes it as a borderline case for Fable 5's safety classifier rather than a fundamentally new capability.
- In response, Anthropic partnered with Amazon, Microsoft, Google, and other Glasswing program members to develop a shared industry framework for assessing and classifying jailbreak severity. The framework aims to standardize how AI developers triage new findings and communicate risk levels to government partners.
- Fable 5 will be available on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Pro, Max, Team, and select Enterprise plans get up to 50% of weekly usage limits through July 7, then it shifts to usage credits.
- Mythos 5 access has been restored for a set of US organizations following government approval on June 26, with ongoing coordination to expand access under Project Glasswing.
- Anthropic also published details on its safety classifier approach, including how it detects potentially dangerous cybersecurity uses and the updated safeguards applied to Fable 5 before redeployment.
Why it matters
This is the first time a frontier AI model has been pulled from global availability and then restored due to cybersecurity capability concerns. The incident sets a precedent for how governments may intervene in model deployment when offensive cyber capabilities are involved. The new jailbreak severity framework could become the industry standard for communicating model safety — similar to how CVE scoring standardized vulnerability severity.
What to do
- If you run Claude models in production, review the updated safety classifier documentation and test your guardrails against the new Fable 5 deployment.
- Track the industry jailbreak framework as it develops — it may affect your compliance requirements and model selection criteria.
- For teams using Mythos 5 through Project Glasswing, coordinate with Anthropic on the expanded access process.
- Document your model deployment decisions: the export control precedent shows that model availability can change on short notice based on government assessment of cybersecurity capabilities.