US Export Directive Forces Anthropic to Take Down Advanced Dual-Use AI Models

Anthropic took its Claude Fable 5 and Mythos 5 AI models offline after a US government directive barred foreign nationals from using them, and the company is still negotiating with the White House to reinstate the services. The Mythos family, which debuted in April, is designed to both detect and exploit software vulnerabilities, making it a powerful but dangerous tool if misused. Anthropic initially restricted access to a select consortium through Project Glasswing, and while Mythos 5 remained private, Claude Fable 5 was released publicly with safeguards blocking responses on biology and cybersecurity.