Anthropic Fable 5 Model Access Restored After Export Control Lifted
Anthropic's Fable 5 and Mythos 5 models had their access restricted due to US export controls, which have now been lifted. Fable 5 will be globally available starting July 1st, with specific access terms for different plans and platforms. The company is also enhancing its safeguards and collaborating on an industry framework for AI model safety.
- →Global availability of Fable 5 begins July 1st
- →Fable 5 and Mythos 5 access restored post-export control
- →Improved safeguards implemented for Fable 5
- →Strengthened collaboration with the US government on AI safety
- →Analysis of Fable 5 vulnerability report and model comparisons
Features (1) ›
- Global availability of Fable 5 begins July 1st
Fable 5 will be available globally on Claude Platform, Claude.ai, Claude Code, and Claude Cowork starting July 1st. Pro, Max, Team, and select Enterprise plans will include Fable 5 for up to 50% of weekly usage limits until July 7, after which it will be available via usage credits.
Enhancements (2) ›
- Improved safeguards implemented for Fable 5
Anthropic has developed an improved safety classifier for Fable 5 that blocks a specific, reported bypass technique in over 99% of cases, addressing concerns raised by Amazon researchers. This enhancement aims to prevent potentially harmful cybersecurity uses while minimizing impact on legitimate coding and debugging tasks.
- Strengthened collaboration with the US government on AI safety
Anthropic is increasing its collaboration with the US government, focusing on new pre-release testing protocols, information sharing initiatives, and joint research efforts to enhance AI safety and security.
Notes (2) ›
- Fable 5 and Mythos 5 access restored post-export control
Access to Anthropic's Fable 5 and Mythos 5 models, initially restricted due to US export controls on June 12, has been restored as of June 30 following the lifting of these controls. Mythos 5 access is currently limited to select US organizations.
- Analysis of Fable 5 vulnerability report and model comparisons
Anthropic's internal testing confirmed that the reported method for bypassing Fable 5's safeguards to identify software vulnerabilities was also exploitable by several other models, including older versions of Claude and GPT. The demonstration of exploiting the vulnerability was reproducible across multiple tested models, indicating the issue was not unique to Fable 5's advanced capabilities.
https://www.anthropic.com/news/redeploying-fable-5