A line of text timestamped 5:21 pm ET on Friday arrived at Anthropic and within hours its newest models were offline worldwide. The company says it disabled access to Claude Fable 5 and Mythos 5 after receiving a U.S. government export control directive citing national security concerns, and it pulled the models for all customers to ensure compliance. Anthropic said the order required it to suspend access by any foreign national, whether inside or outside the United States, and that it couldn't enforce a selective block in real time. The firm apologised for the disruption and said all other Claude models would remain available while it engages with authorities.

Inside Anthropic's offices, engineers and product managers saw the same short instruction and the same problem: a government letter sent at 5:21 pm ET asking the company to stop foreign nationals from using two newly released models, Claude Fable 5 and Mythos 5, and no reliable way to apply that restriction in real time.

Anthropic said in a blog post that it received a U.S. government export-control directive that cited national security concerns but didn't include technical details. The company said the directive asked it to block access by "any foreign national, whether inside or outside the United States, including foreign national Anthropic employees." Faced with a requirement it couldn't implement selectively and quickly, Anthropic removed access to Claude Fable 5 and Mythos 5 for every customer rather than attempting piecemeal enforcement.

The takedown was immediate and global. Anthropic apologised to customers for the disruption and emphasised that all other Claude models would remain available. The company added that it had reviewed a demonstration of the alleged bypass method and judged that it identified a small number of previously known, minor vulnerabilities. Anthropic asserted those issues appeared simple and said other publicly available models could discover them without a bypass.

How the models reached this point

Claude Fable 5 was released to the public on 9 June as a general-availability iteration of Anthropic's Mythos class of models. The company positioned Fable 5 as its most capable public offering to date and said it shipped with new safeguards designed to stop it answering certain high risk queries, including on cybersecurity, biology and chemistry. The preceding Mythos Preview model had been rolled out to a small group of organisations in April with the stated aim of giving defenders tools to find and fix vulnerabilities.

Mastheads

Technical reporting around the launch underlined why the models drew scrutiny. Commentators noted Fable 5's unusually large prompt context and higher token consumption compared with Anthropic's Opus series.

Anthropic also said it had implemented pre-release collaborations and safeguards as part of a limited rollout to select organisations, and that internal initiatives accompanied the preview phase.

Anthropic has publicly disputed the government's assessment. The company characterised the alleged jailbreak as narrow and not materially increasing an attacker’s capabilities compared with other models. In its blog post Anthropic said the government provided only verbal evidence of the reported bypass, and the firm objected to the lack of technical detail in the directive.

Those objections sit against a backdrop of growing tension between Anthropic and US authorities. Earlier this year the U.S. Department of Defense labelled Anthropic a "supply chain risk," a designation the company contested and which has barred US government agencies and contractors from using Anthropic technology. Anthropic has filed lawsuits challenging that designation and has argued that any authority to block deployments should follow a transparent statutory process that's fair, clear and grounded in technical facts.

For customers the practical consequence is immediate. Organisations that had begun integrating Fable 5 or Mythos 5 into products or services must now revert to other Claude models or alternative systems while Anthropic engages with the government. The company said it couldn't reliably enforce a geographic or identity based block in real time, which is why it chose to disable the two models worldwide rather than risk non compliance.

The dispute highlights the frictions that can arise as advanced models move from limited previews to broad availability. Anthropic argued it had layered in safeguards and limited rollouts to reduce risk, and it said a demonstration of the alleged bypass revealed only minor, previously known issues. The government, according to the company, sees the same technique as a national security concern.

Legal and regulatory experts will watch how the two sides resolve their differences. Anthropic has asked for more technical information from the authorities and continues to press that any restrictions be set through a clear legal process. For now the decision prompted a blunt operational fix: two models taken offline until the company and the government find a path that addresses the security concern while permitting legitimate use.

Related Articles

The 5:21 pm ET letter set in motion a legal and technical standoff. Anthropic is engaging with authorities and pursuing its remedies; the next milestones to watch are whether US officials provide technical detail or a revised directive, and any court rulings that could restore access to the affected models.

This article was created with AI assistance.