US lifts export controls on Anthropic’s Mythos 5 and Fable 5 AI models

The decision concludes a nearly three-week restriction initiated over national security concerns regarding potential vulnerabilities to cyberattacks.

Access to the more specialized Mythos 5 model has been reinstated specifically for organizations involved in Project Glasswing, a consortium of over 100 companies and government agencies tasked with operating and defending critical infrastructure.

The models were initially blocked by an export control directive on June 12, following reports from Amazon researchers who identified a method to bypass Fable 5’s safety guardrails. In a June 26 letter confirming the reversal of the directive, Secretary of Commerce Howard Lutnick acknowledged Anthropic's subsequent cooperation with federal authorities and Amazon. The company has since implemented an improved safety classifier and established new protocols for reporting malicious activity.

According to statements from Anthropic, the vulnerability that prompted the restriction was not isolated to Fable 5. 5 and Kimi K2.7, possessed similar flaws. Anthropic noted that the initial global suspension was necessitated by technical limitations in verifying user nationality in real-time under US jurisdiction.

The rapid implementation and subsequent lifting of these controls illustrate the structural difficulties of regulating advanced AI models through unilateral national mechanisms. As similar vulnerabilities appear concurrently across multiple global platforms, the episode indicates that managing AI security risks and critical infrastructure access will increasingly demand coordinated international regulatory frameworks rather than localized export restrictions.