The US government has partially lifted restrictions on Anthropic's most advanced AI model, Claude Mythos 5, clearing the company to share it with a limited set of approved users.
According to MSN's reporting, the move follows initial concerns that the model could be misused by adversaries. CXOToday reports that under the Trump administration's decision, Anthropic may distribute Mythos 5 only to "trusted partners" rather than releasing it broadly.
The outcome amounts to a notable shift. As Storyboard18 notes, Anthropic had held back Claude Mythos — described as its most powerful model yet — before its release. Inc.com frames the government's decision as a "dramatic reversal" that opens access to that model, but with strings attached. Digit reports that Anthropic restored limited access to Mythos 5 for approved users after talks with the US government.
The loosening is not total. Digit reports that restrictions on a separate model, Claude Fable 5, remain in place. Other outlets, including a report cited from Azerbaijan, similarly describe access being granted only to "select groups."
The episode highlights a broader policy puzzle. The Bulletin of the Atomic Scientists argues that AI models like Claude Fable and Mythos defy traditional export control frameworks — the rules historically built to govern physical goods and weapons technology rather than software that can be copied and distributed.
The sources do not specify which organizations qualify as trusted partners, the exact criteria, or the timeline involved.
Why it matters: the decision shows governments are now treating the most capable AI systems like controlled strategic technology, weighing commercial release against national-security risk.