US lifts export restrictions on Anthropic AI models

The U.S. Treasury Department has removed export controls on Anthropic’s most advanced language models, Claude Fable 5 and Claude Mythos 5, allowing the company to restore worldwide access. The models had been blocked in June after the Trump administration flagged security risks, citing vulnerabilities that could enable misuse in cyber‑attacks. Anthropic announced it would resume global availability and continue cooperating with the government, noting that a limited group of U.S. cybersecurity firms already received access under a special permit.

The policy shift follows a broader U.S. effort to establish a voluntary framework for testing and monitoring powerful AI systems, a move mirrored by other firms such as OpenAI, which also face export limits on its upcoming GPT‑5 model. Anthropic called for standardised guidelines to assess critical weaknesses in advanced models as they become increasingly capable of affecting software security worldwide.