Some early users of Anthropic's Mythos AI model retain access to a preview version despite a U.S. government order that led to the shutdown of other versions. The specific criteria for continued access remain unclear.
The continued access for select users of Anthropic's Mythos AI model, despite a U.S. government shutdown order and national security concerns, raises questions about the enforcement and scope of export controls on advanced AI technologies.
Despite a U.S. government order that led to the shutdown of certain versions of Anthropic's Mythos AI model, some early users selected for its Glasswing program continue to have access to a preview.
Anthropic had initially limited the Mythos Preview to approximately 200 organizations, including U.S. government entities, after the model identified numerous software vulnerabilities. While a less powerful version was released publicly and subsequently disabled due to national security concerns and Washington's export control directives, firms such as Dragos and Cisco Systems have confirmed they still have access to the preview.
The criteria Anthropic used to determine which Glasswing members retained access were not immediately clear. The European cybersecurity agency Enisa, initially invited to the program, was informed that its access would be revoked.
The relationship between Anthropic and the U.S. government reportedly soured earlier this year when the AI company refused to permit the U.S. military to utilize its models for domestic surveillance and fully autonomous weapons systems.