US Order Forces Anthropic to Kill Its Top AI Models

Anthropic disabled Fable 5 and Mythos 5 for all customers after a U.S. export-control order barred access by any foreign national, including Anthropic’s own employees. That is the clearest signal yet that Washington is treating frontier AI models as national-security assets, not ordinary cloud software.

The order landed Friday, according to Time, and Anthropic said it had to “abruptly disable” access to Fable 5 and Mythos 5 for all customers to comply. The directive was aimed at foreign nationals, but the company’s practical response was broader: shut the models off entirely while it contests the government’s rationale.

“We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible,” Anthropic said.

Anthropic’s pullback turns frontier models into controlled infrastructure

The policy turn is sharper than the outage. The U.S. has used export controls before to restrict sales of the semiconductor chips that power AI systems. Time reports that this is the first time those controls have been applied to the models themselves.

That changes the control point. Washington is no longer just trying to shape who can buy the hardware used to train frontier systems. It is moving toward rules on who can touch the finished capability.

XOOMAR analysis: This is the deeper shift beneath the headline. Advanced AI access is becoming a national-security checkpoint. Anthropic built a global product. The U.S. government treated that product as something that may need nationality-based restriction, even when access runs through a commercial AI company.

That puts labs in a bind. If the government can force access changes at the model level, frontier AI companies have to design for policy interruption as much as performance. For product context on the launch, see XOOMAR’s earlier analysis of Claude Fable 5 selling Mythos-class AI on a short leash.

A jailbreak claim triggered a shutdown larger than the stated target

The stated concern was a possible jailbreak in Fable 5. Anthropic said the government notified it Friday that it had become aware of a method of bypassing safeguards. The company pushed back, saying it had received only verbal notice of a “potential narrow, non-universal jailbreak.”

The disputed capability matters because Anthropic said at launch that Fable 5 was especially strong at identifying software vulnerabilities. In its June 9 release statement, the company warned that “[r]eleasing a model this capable comes with risks,” and said safeguards were added to block use in certain areas.

Anthropic’s own wording gave regulators a handle:

“Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage,” the company said.

The government appears to be arguing that if the safeguards can be bypassed, the model’s cyber capability becomes an export-control problem. Anthropic argues the evidence does not justify a recall-like shutdown.

For readers tracking the immediate access fallout, XOOMAR has a related breakdown of the U.S. order knocking Claude Fable 5 offline after jailbreak fear.

The market read the order as an IPO risk

The timing is brutal for Anthropic. Time reports that the announcement came ahead of an expected U.S. IPO, while SpaceX launched its own IPO on Friday and became the sixth-most-valuable public company in the U.S., with a $2.1 trillion market cap. OpenAI is also reportedly considering a similar move.

The reaction in synthetic pre-IPO markets was immediate. CoinDesk reported that the Anthropic perpetual futures contract on Hyperliquid dropped 3.7% on Saturday to about $1,627, after trading near all-time highs above $1,800 in the days after Fable 5 launched. Open interest was near $8.6 million.

That is not a public equity market verdict. Anthropic has not completed an IPO. But it shows how traders framed the issue: not as a temporary product outage, but as a new regulatory discount on the company’s most valuable systems.

XOOMAR analysis: The risk is not just lost access for a weekend. It is uncertainty over whether Anthropic can commercialize its most capable models globally if Washington can narrow, suspend, or condition access on short notice.

Safety branding now cuts both ways

Anthropic has built its identity around safety-conscious AI development. That usually helps with regulators, enterprise buyers, and public trust. Here, it may have raised the stakes.

The company itself described Fable 5 as exceeding “those of any model we’ve ever made generally available.” It also acknowledged cybersecurity misuse risk. Those statements are part of responsible disclosure, but they also make it easier for officials to argue that the product sits in a national-security category.

The Pentagon’s Chief Information Officer, Kirsten Davies, framed the decision in hard political terms on X:

“Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuation. America First. Always,” she wrote.

Anthropic’s dispute with the Trump Administration adds another layer. Time reports that the clash began when the company refused to allow the U.S. military to use its AI models for fully autonomous weapons systems. The Pentagon then placed Anthropic on a blacklist, deeming it too dangerous for government use. Now the models have also been treated as too dangerous for foreign use.

That is an awkward position for a safety-first lab: distrusted by parts of government for refusing certain military use, then restricted by government because its models may be too powerful to export broadly.

Foreign customers and allies got a sovereignty warning

The foreign-access ban does not only affect adversaries in theory. Time says the order covered any foreign national, including Anthropic employees. That breadth is why the decision immediately spilled into the AI sovereignty debate.

Anton Leicht, a fellow with the Technology and International Affairs Program at the Carnegie Endowment for International Peace, told Time the shutdown reveals how little foreign market access seemed to weigh in the decision.

“It shows how irrelevant most other countries have become to AI policy. It seems like neither access to foreign markets nor any retaliation options held by any other country factored into the administration's decision,” he said.

British lawmaker Kanishka Narayan, minister for AI and Online Safety, said the ban should push deeper investment in the U.K.’s own AI industry.

Leicht was skeptical that sovereignty rhetoric can close the gap quickly. He told Time: “Only the U.S. builds frontier models, and the U.S. controls almost all the chips needed to train them.”

That is the hard lesson for allies and customers. Access to top U.S. AI models can be cut by U.S. policy, even if the customer is legitimate, commercial, and outside the immediate dispute.

The next fight is the evidence standard for model shutdowns

The unresolved question is not whether AI models can pose security risks. Anthropic itself says Fable 5’s cybersecurity capability could cause serious damage without safeguards. The live dispute is what level of proof should justify pulling a commercial frontier model from global access.

Anthropic says the government has provided only verbal evidence of a narrow jailbreak. The company argues that treating such a finding as enough to suspend deployment could halt new model releases across the frontier AI sector if applied consistently.

That is where the next phase turns. Watch for three signals:

• Evidence: Whether the government provides technical details that support a broad national-security action.
• Scope: Whether access is restored, narrowed to certain users, or extended to other high-capability models.
• Policy design: Whether future frontier AI releases include stricter access controls before regulators intervene.

The practical takeaway is stark. The most powerful AI models are no longer just products competing on benchmarks. They are becoming controlled capabilities, and access may depend as much on security policy as model performance.

Impact Analysis

• The order signals that frontier AI models may now be regulated like strategic national-security assets.
• Anthropic’s full shutdown shows how nationality-based rules can disrupt commercial AI products for all customers.
• AI labs may need to design products around export-control compliance as much as technical performance.