The US government issued an emergency directive in mid-June 2026, temporarily suspending access to Anthropic's new Claude Mythos AI model. The action cited national security concerns tied to a reported jailbreak vulnerability.
The suspension is significant because Mythos was deployed under a highly controlled program called Project Glasswing. Its initial release in April 2026 was capped at about 200 vetted organizations, including US government entities, and followed a voluntary 30-day federal review.
Anthropic had highlighted Mythos's capability to identify thousands of software vulnerabilities, including zero-day exploits. The company also committed up to $100 million in usage credits for participating organizations.
This incident occurs despite major cloud partnerships with Amazon Web Services, Microsoft, and Google, and raises the implied risk for less carefully managed AI deployments across the broader market.