Anthropic Restores Claude Fable 5 After US Lifts Export Controls
In brief
- Claude Fable 5 and Mythos 5 restored to public access after US lifts export controls
- Models restricted since June 12 following researcher safeguard bypass report
- Anthropic deploys new classifiers to block cybersecurity-related tasks
- Commerce Secretary Lutnick and White House approved redeployment
Export Controls and Initial Restrictions
The export controls were applied after Amazon researchers found a method of bypassing Fable 5's safeguards, prompting the model to identify software vulnerabilities. The incident triggered government concern about frontier AI technology and state oversight. An AI researcher claimed to have jailbroken Fable 5 within 48 hours of its launch in June, adding pressure on regulators to act.
Anthropic argued the vulnerability wasn't unique to Fable 5. Weaker models could also identify the same vulnerabilities, the company said. Still, the suspension raised concerns about state control over frontier AI and set a precedent for how government could regulate AI development.
Government Approval and Redeployment
US Secretary of Commerce Howard Lutnick stated the government worked closely with Anthropic to analyze and approve Fable 5. The approval came after what Anthropic described as productive conversations with officials.
"After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks." — Anthropic
The new classifiers represent the core of Anthropic's strategy to address government concerns while maintaining access to its frontier models. White House Chief of Staff Susie Wiles stated the government priority is to deploy the best AI technology as quickly and safely as possible.
Collaborative Safeguards
Anthropic is collaborating with Amazon, Microsoft, Google and other partners in Project Glasswing to safeguard against AI cybersecurity threats. The framework aims to build consensus around how frontier AI labs should handle security-related capabilities. This multi-company approach signals a broader shift toward industry-wide standards for managing AI risks.
