The US directive to prevent foreign access to Anthropic's leading models signifies a shift in policy.

      The US government has instructed Anthropic to prevent foreign nationals from utilizing its two most advanced AI models. Instead of trying to implement a nationality-based restriction selectively across a shared cloud service, the company opted to disable the models for all users.

      On Friday, June 12, Anthropic shut down Claude Fable 5 and Claude Mythos 5 globally, just three days after introducing Fable 5 as its most powerful public offering. This action is reportedly the first export-control measure focused specifically on certain AI models, rather than the chips or hardware that support them. The directive prohibited access for foreign nationals both within and outside the United States, making selective enforcement across a multi-tenant service impractical and leading to a complete shutdown as the easiest solution.

      Anthropic understands the government’s concern revolves around a method to jailbreak Fable 5, allowing users to bypass the safeguards intended to prevent the model from generating harmful outputs. This action followed a jailbreak demonstration published on X on June 10 by a well-known figure who claimed to have circumvented the model’s safety measures.

      Anthropic reviewed the report believed to have prompted the government directive and concluded that the capability demonstrated is readily available in other models, including OpenAI’s GPT-5.5. The company described the response as excessive, arguing that recalling a commercial model used by hundreds of millions of users due to a narrow potential jailbreak could “halt all new model deployments for all frontier model providers” if this approach became standard practice.

      This situation extends beyond Anthropic: it sets a precedent where the government can withdraw a launched model for national security reasons, affecting the entire industry rather than just one company. The jailbreak that sparked the controversy was publicly shared on X on June 10 by a prominent individual in the model-breaking community, who claimed to have breached Fable 5’s safety controls.

      The government’s order came within days, which is part of what Anthropic contests: a publicly available demonstration was used as the basis to withdraw a model from hundreds of millions of users, rather than a private investigation.

      For the administration, this action aligns with a more proactive stance towards an industry it has otherwise sought to engage, and the decision to intervene with a flagship product shortly after its release signals a new approach. The method of enforcement is also significant. Since the directive targets foreign nationals using a shared cloud service, compliance necessitated turning off the models for everyone, transforming a nationality-specific restriction into a global outage.

      For enterprise clients who had integrated Fable 5 and Mythos 5 into their operations, the immediate issue is quite tangible: the models they relied on are now inactive, and they do not have control over when they might return. The next steps are contingent on developments in Washington.