US Bans Fable 5 Release Due to Alleged National Security Concerns

The US government issued an export control directive on June 12, 2026, ordering Anthropic to suspend all access to its two newest AI models, Claude Fable 5 and Claude Mythos 5, for any foreign national inside or outside the United States . Because separating foreign nationals from other users was technically infeasible, Anthropic disabled both models for all customers globally . The move marked a striking use of export authority against an American AI lab's own products.

The stated technical trigger was a guardrail bypass discovered by Amazon researchers: the difference between asking the model to "review code for security issues" versus asking it to "fix this code." Anthropic reviewed a demonstration of the technique and concluded it exposed only a narrow, non-universal set of previously known minor vulnerabilities . The company stated it had not received disclosure of any harmful result from the jailbreak and argued the same capability is routinely used by security defenders across the industry and is available in competing models .

Cybersecurity experts and researchers pushed back sharply. Katie Moussouris and others signed an open letter arguing the jailbreak in question should never have triggered an export control, and that patching it would actually weaken the model's ability to assist defensive security work . TechCrunch reporting from June 15 suggests the ban may have had more to do with political friction between the Trump administration and Anthropic than with the technical threat itself, citing personality differences between the administration and Anthropic leadership and possible pressure through Amazon CEO Andy Jassy .

Despite the disruption, early indicators suggest Anthropic's brand may not have suffered lasting damage. The forced withdrawal drew enormous public attention to Fable 5 and Mythos 5 and highlighted their capabilities, potentially driving interest once the models are restored. The episode nonetheless sets a precedent that the US government can unilaterally force a domestic AI company to take models offline without court approval, a power with significant implications for every frontier model provider going forward .