Anthropic has been compelled to restrict access to its latest AI models after receiving a direct order from US government agencies citing national security concerns. The company confirmed in a Friday blog post that it blocked all foreign nationals—including those physically present in the US and employed at Anthropic—from using Fable 5 and Mythos 5, effective immediately at 5:21 PM ET.

The directive arrived without detailed justification, leaving Anthropic to piece together the rationale. Internal analysis suggests the trigger was a reported capability in Mythos 5 for reviewing and correcting software code, including vulnerabilities that had remained latent for decades. Anthropic’s engineers note that this functionality is not unique—OpenAI’s GPT-5.5 offers similar code analysis—and that Fable 5’s safety layers were extensively validated before release. The company disputes that a feature available in competing models warrants a blanket access ban for hundreds of millions of users.

Architecturally, Mythos 5 remains a non-public, full-capability model reserved for government and select corporate partners to harden systems. Fable 5, released this week as a derivative, has its cybersecurity and biotechnology modules disabled. The order also impacts foreign researchers and staff currently in the US, forcing Anthropic to implement an abrupt revocation of API keys and user sessions with no transition period.

Anthropic’s leadership has previously called for coordinated industry pauses on advanced AI development, warning that rapid capability growth risks loss of human control. This incident underscores the tension between open research and national security—and the operational complexity of enforcing geopolitical boundaries on globally distributed model access.

Source: Al Jazeera