US lifts export controls on Anthropic’s Claude Fable 5, global access resumes

The US government has lifted export controls on Anthropic's Claude Fable 5 and Claude Mythos 5, enabling the company to restore global access to its latest AI models after a three-week restriction. Following concerns raised by an Amazon research report regarding potential cybersecurity vulnerabilities, the US government had imposed controls that forced Anthropic to suspend access due to inability to verify user nationality. Access is scheduled to resume globally via the Claude platform and Claude.ai starting July 1.

To address the security concerns, Anthropic has implemented enhanced safeguards, including a new safety classifier designed to block jailbreak techniques with a 99% success rate. While this may increase false positives during coding tasks, the company is working to refine the system. Anthropic is also collaborating with US government agencies to deepen cooperation through pre-release evaluations, information sharing, and joint research.

Furthermore, Anthropic is leading an initiative with partners like Amazon, Microsoft, and Google to develop a standardized industry framework for assessing AI jailbreaks. This framework will evaluate capability gain, breadth of capability, ease of weaponization, and discoverability to help governments and companies respond to emerging AI security threats more effectively.

(Source：Medianama)