Anthropic restored access to Claude Fable 5 on 18 June, ending the six-day shutdown triggered by a US Commerce Department export control directive on 12 June. However, the model that returned is not the same model that went offline — Anthropic implemented several significant changes as conditions of restoration.
The restored Fable 5 includes nationality-based access controls that restrict usage based on the user's verified location, mandatory identity verification for API access in certain jurisdictions, tighter safety classifiers that activate more frequently than the original deployment, and mandatory data retention requirements. These changes represent a meaningful shift from Anthropic's pre-shutdown stance, where CEO Dario Amodei had initially refused the administration's demand to either fix the jailbreak vulnerability or voluntarily de-deploy the model.
The shutdown had prompted an unprecedented response from the cybersecurity community. Over 300 leaders — including Alex Stamos, Bruce Schneier, and executives from Google, NVIDIA, and Adobe — signed an open letter at freefable.org arguing that 'to pull the best capabilities away from defenders without a good reason when our adversaries are rapidly advancing is dangerous.' The letter framed the ban as counterproductive to national security rather than supportive of it.
Anthropic's Chris Ciauri, speaking at the company's Seoul office opening that coincided with restoration, expressed confidence that full access would be maintained going forward. A more comprehensive identity verification system is expected to launch around 8 July, which would further refine access controls based on user verification status.
For context engineers, the Fable 5 restoration with new restrictions establishes a precedent: frontier AI models can be recalled and modified by government directive, and the restored versions may include access controls and monitoring that did not exist in the original release. This has significant implications for production systems built on specific model behaviours.
