Some early participants in Anthropic’s Project Glasswing can still access the preview version of its AI model Mythos even after a prior U.S. government-related ban called for a full shutdown, Bloomberg reported. The testing program includes about 200 approved organizations, including banks and technology companies, that use the model for cybersecurity vulnerability detection and research. Dragos and Cisco said they still have access. The development suggests the shutdown has not been applied uniformly across all previously selected users, leaving parts of the pilot program active for security-focused work.