Anthropic blocks flagship models after US government order

An export control order has abruptly redrawn access to frontier AI

Anthropic says it has cut off access to its Fable 5 and Mythos 5 models after a US government export control directive ordered the company to block the systems for all foreign nations and foreign nationals, whether they were inside or outside the United States. According to the supplied report, the order also covered Anthropic employees, prompting the company to fully disable customer access in order to comply.

This is a consequential step not just for Anthropic, but for the governance of advanced AI models more broadly. Export controls have long been a tool for managing sensitive technologies, but applying them directly to model access at this level raises new questions about what governments believe frontier systems can enable and how quickly access restrictions can be imposed when security concerns are invoked.

The dispute is over both access and evidence

Anthropic says it is complying, but it is also openly contesting the basis for the action. In the statement described by The Verge, the company said the government did not provide specific details of its national security concern. Anthropic further claimed that any evidence of a possible jailbreak was delivered verbally and that the vulnerabilities identified were minor and available through other models, including GPT 5.5.

That matters because the policy significance of the order depends not only on what the models can do, but on the standard of evidence required to justify broad restrictions. If access can be halted across entire populations without detailed disclosure, model providers may be forced into compliance while still lacking a clear technical case they can independently assess or remedy.

Anthropic also used the statement to defend its own safeguards, saying it had worked with the US and UK governments and modified its data retention policy to better track potentially malicious use. The company argued that it had not received disclosure of a non-universal jailbreak that led to a harmful result and that the issues raised did not provide a Mythos-specific advantage.

A new phase in AI policy

The immediate effect is straightforward: customers lost access. The broader effect is that frontier AI regulation appears to be moving closer to an export-control framework traditionally associated with chips, cryptography, aerospace, and dual-use technology. That is a major escalation in how model capability is being treated by the state.

The order also highlights a hard operational problem for AI companies with global workforces and international customer bases. The report says the directive applied to foreign nationals even within the US and included Anthropic employees. That makes compliance less like geoblocking a region and more like reclassifying who is allowed to touch the product at all.

For developers and enterprises, the message is equally stark. Access to a leading model may no longer depend only on pricing, technical integration, or vendor preference. It may depend on nationality, residency, and rapidly changing government directives that arrive with limited public explanation.

What this could mean next

Even with sparse official detail, the event is likely to reverberate across the AI sector. Rival companies will have to consider whether similar orders could reach their own top-tier systems. Enterprise customers will be pushed to think more seriously about concentration risk if critical model access can disappear overnight. And policymakers will face pressure to explain where the line sits between prudent national security controls and opaque, disruptive intervention.

Anthropic's statement makes clear that the company does not accept the government's implied technical case, even as it obeys the directive. That tension may define the next stage of frontier AI policy: firms building powerful systems under growing pressure to show they can police misuse, and governments increasingly willing to act first and explain later.

For now, the practical reality is simple. Two advanced models have been cut off under an export control order, and the terms of access to frontier AI have become visibly more political, more conditional, and less stable than they appeared even a week ago.

This article is based on reporting by The Verge. Read the original article.