The U.S. Commerce Department placed Anthropic’s Claude Fable 5 and Mythos 5 under export controls on June 12, leading Anthropic to disable both models for every customer worldwide within hours, according to the provided source material and Anthropic’s account. The order matters because it showed that a commercial frontier AI service can be switched off globally when U.S. national-security officials decide access poses a risk.

Commerce Secretary Howard Lutnick sent the order to Anthropic CEO Dario Amodei at 5:21 p.m. ET on June 12, three days after Anthropic released the Mythos-class models. The order barred access for any foreign national, anywhere, including foreign-national employees inside Anthropic, according to the source material. Anthropic said there was no clean way to comply with that scope while keeping the models online, so it disabled Claude Fable 5 and Mythos 5 for all users by midnight.

Fable 5 was the public, heavily guarded commercial model. Mythos 5 was described as the more powerful underlying system, routed to selected organizations for cyber-defense work through a program called Project Glasswing rather than released openly. Anthropic said the order cited national-security authorities but gave no specific rationale, and the company publicly called the action a misunderstanding.

Confirmed: the models were launched June 9, placed under export controls June 12 and disabled worldwide the same night. Reported or claimed: officials and outside testers raised concerns about jailbreaks, cyberattack assistance and possible access by a China-linked group. Still unresolved: whether the reported security failures were severe enough to justify a global shutdown.

Reliability Risk Hits AI Buyers

The shutdown changes the risk calculation for companies building products, security workflows or research pipelines on hosted frontier models. Until now, buyers often treated access outages as operational problems. The Anthropic order adds a regulatory risk: a model may be technically available, commercially licensed and then removed because of government action.

That risk could push customers toward multi-model contracts, sovereign AI systems and open-weight models that can be hosted outside a single vendor’s control. The source material cites Deutsche Bank’s reported buyer concern as Can’t rely on it, a short phrase that captures the commercial damage Anthropic and other U.S. labs may face if customers see hosted frontier AI as revocable infrastructure.

The timing also matters for capital markets. The source material says the dispute lands weeks before both labs are expected to go public. Any effect on IPO plans, valuations or investor demand remains uncertain and is not financial, tax or legal advice.

From Launch To Shutdown

Anthropic released the Mythos-class models on June 9 and marketed them for advanced cybersecurity and biomedical use. By June 12, the models had shifted from product launch to national-security case. The source material frames the episode as the first time the U.S. government turned off access to a frontier AI model through export controls.

The policy dispute sits inside a broader argument over whether export rules built for chips, hardware and physical goods can fit software services that are already deployed across global cloud systems. Anthropic and more than 120 cybersecurity executives and engineers argue that removing the models from defenders may also weaken organizations using them for security work.

The government-side case, as described in the source material, rests on concerns about malicious outputs, agentic cyber tasks, proliferation and reverse-engineering. The industry-side response is that comparable capabilities may exist in other U.S. and Chinese systems, so blocking one hosted model may not remove the underlying capability from the world.

“misunderstanding”

— Anthropic

Evidence Behind The Ban Disputed

It is not yet clear which finding drove Commerce to act, or whether one report, several reports or wider intelligence concerns led to the order. The U.K. AI Safety Institute’s red-team lead said publicly that his team built a jailbreak within hours and later extended it to multi-step agentic tool calls. The Wall Street Journal, according to the source material, reported that Amazon warned officials Fable 5 could provide information usable in cyberattacks.

Semafor reported that suspicion of access by a China-linked group also played a role. Anthropic disputes that the known jailbreak evidence justified recalling a model already deployed to a large user base, saying the model had passed thousands of hours of red-team work by internal teams, government testers and third parties without a universal jailbreak being found.

June 22 Talks Ahead

Anthropic and the White House are scheduled to meet June 22. The immediate questions are whether Commerce narrows, lifts or maintains the export controls, and whether Anthropic can propose technical limits that satisfy officials without keeping both models offline.

For the wider AI industry, the next milestone is policy clarity. Labs, cloud providers and enterprise buyers will be watching for whether this order remains a one-off response to specific model concerns or becomes a template for future frontier AI controls.

Key Questions

What did the U.S. order Anthropic to do?

The order placed Claude Fable 5 and Mythos 5 under export controls and barred access by foreign nationals, according to the source material. Anthropic said that scope made selective compliance impractical, so it disabled both models worldwide.

Why were the models shut down for U.S. customers too?

Anthropic’s position is that the order covered access conditions it could not reliably separate across its live service, including foreign-national employees and users. The company chose a global shutdown to comply within hours.

Was the reported jailbreak confirmed?

The source material cites public and media reports of jailbreak concerns, including work by the U.K. AI Safety Institute and Amazon researchers. Anthropic says any known jailbreak was narrow and not a universal break of the model’s protections.

Does this affect other AI companies?

Yes, even if the order applies to Anthropic’s models. The episode gives buyers a concrete example of government action cutting off access to a hosted frontier model, which may affect procurement, risk planning and model diversification.

When could the models return?

No return date has been announced. The next known step is a June 22 meeting between Anthropic and the White House, where the company is expected to argue for the controls to be lifted or narrowed.

Source: Thorsten Meyer AI