TL;DR
The U.S. Commerce Department placed Anthropic’s Claude Fable 5 and Mythos 5 under export controls on June 12, forcing the company to disable both models for all customers because it could not limit access only by nationality. The order has become a test case for whether frontier AI can be treated like export-controlled goods, while buyers weigh the risk that cloud-hosted models can be cut off fast.
The U.S. Commerce Department placed Anthropic’s Claude Fable 5 and Mythos 5 models under export controls on June 12, forcing the company to disable both systems worldwide and turning a major AI product launch into a national-security dispute with broad consequences for enterprise buyers, cloud providers and AI investors.
According to Anthropic’s account, Commerce Secretary Howard Lutnick sent CEO Dario Amodei a letter at 5:21 p.m. ET on June 12 barring access to the models by any foreign national, including foreign-national employees inside Anthropic. The company said it could not cleanly comply with that condition on a user-by-user basis, so it disabled the models for every customer by midnight.
The order came three days after Anthropic released the Mythos-class models on June 9. Fable 5 was the public commercial version, while Mythos 5 was a more powerful underlying model made available to selected organizations for cyber-defense work under Project Glasswing, according to the source material. Anthropic said the government cited national-security authorities but did not provide a detailed public rationale.
The reported basis for the order remains contested. The U.K. AI Safety Institute’s red-team lead said his team built a jailbreak that produced malicious answers and later extended it to agentic tool calls. The Wall Street Journal reported that Commerce was alarmed by a separate Amazon finding, while Semafor reported that officials also suspected a China-linked group may have obtained access to the model. Anthropic has called the move a “misunderstanding” and said a narrow jailbreak does not justify recalling a deployed model.
Washington just switched off
a frontier model
On June 12, an export-control order forced Anthropic to disable Claude Fable 5 and Mythos 5 worldwide. The security merits are still contested. The lesson buyers took away is not: frontier AI can be turned off.
■ The government’s case
- A reported jailbreak pulled malicious, agentic outputs (UK AISI)
- Amazon told officials Fable yielded cyberattack-usable info
- Suspicion a China-linked group obtained the model
- Proliferation & reverse-engineering risk to national security
▲ Anthropic & 120+ experts
- Calls it a narrow, non-universal jailbreak — a “misunderstanding”
- Capability is real but not unique (GPT-5.5, Opus, Kimi 2.7)
- Controls remove tools from defenders, not just attackers
- Export rules built for chips & ore don’t fit software
The precedent is the story. Whatever the jailbreak’s true severity, the U.S. showed it can dark a commercial American model worldwide on ~90 minutes’ notice. Adoption was supposed to be the moat — this week it became the exposure, and the likely winner is the open, sovereign, self-hosted stack.
Cloud AI Reliability Takes A Hit
The immediate business lesson is that frontier AI access can be interrupted by government order, even when the product is already in customer use. That matters for banks, defense contractors, software firms and health organizations building workflows around hosted models. For those buyers, the issue is no longer only model quality or price, but whether access can disappear during a regulatory fight.
The move may also affect how companies design AI systems. The source material cites Deutsche Bank’s reported concern that buyers “can’t rely on it,” and points to a likely push toward multi-model setups, self-hosted systems and open-weight alternatives. Those choices could weaken the commercial advantage of U.S. frontier labs if customers decide that hosted access carries too much policy risk.
The timing adds pressure because Anthropic and other major labs are expected to seek large public-market valuations. Investors have priced frontier models as scalable global products. A precedent showing that the U.S. can switch off a commercial model worldwide challenges that assumption, even if officials later narrow or reverse the order.
AI Engineering: Building Applications with Foundation Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
A Chip Rule Applied To Software
Export controls have long been used to restrict advanced chips, manufacturing equipment and other strategic technologies. Applying that approach to a deployed AI model creates a harder compliance problem: software can be accessed remotely, incorporated into cloud services and used by employees or customers across borders.
Anthropic said Fable 5 had gone through thousands of hours of red-team testing by internal teams, the U.S. government, the U.K. AI Safety Institute and third parties, and that no universal jailbreak had been found. Critics of the order, including more than 120 cybersecurity executives and engineers who signed an open letter, argue that removing the model also removes tools from defenders.
Supporters of tighter controls argue that cyber-capable models create proliferation risks if hostile actors can use or reverse-engineer them. The available accounts do not establish publicly whether the reported jailbreak, the Amazon finding or suspected China-linked access was the decisive factor.
“misunderstanding”
— Anthropic
The Trigger Is Still Disputed
It is not yet clear which concern drove the Commerce Department’s decision: the reported U.K. jailbreak, Amazon’s reported warning, suspected China-linked access, broader proliferation risk or some combination of those factors. The government has not publicly released the full technical basis for the order.
It is also unclear whether the controls will remain in place, be narrowed to certain users or regions, or be replaced by a patch-and-review process. Anthropic disputes the premise that Fable 5 posed a unique risk, while officials and outside testers have described security concerns that they say required fast action.
June 22 Talks Set The Pace
Anthropic and the White House are scheduled to meet on June 22. That meeting is expected to shape whether Fable 5 and Mythos 5 return, whether access is restored only for certain customers, and whether future frontier model releases face pre-launch export review.
Customers are likely to review contingency plans now rather than wait for a final ruling. The near-term test for the AI industry is whether providers can offer clearer guarantees about access, compliance and shutdown procedures without weakening the security controls that led to the order.
Key Questions
What did the U.S. government order Anthropic to do?
The Commerce Department placed Claude Fable 5 and Mythos 5 under export controls, barring access by foreign nationals. Anthropic said it disabled the models for all users because it could not reliably apply the restriction only to specific people.
Why were the models disabled worldwide?
According to Anthropic, the order applied so broadly that partial compliance was not practical on short notice. The company shut off both models globally within hours.
Was a jailbreak confirmed?
Several reports describe jailbreak findings, including one from the U.K. AI Safety Institute and another involving Amazon. Anthropic says the issue was narrow and not a universal jailbreak. The full technical record has not been made public.
Why does this matter for AI customers?
It shows that access to a hosted frontier model can be interrupted by government action. Companies using such models may now treat regulatory shutdown risk as part of vendor planning.
When could the situation change?
The next clear milestone is the June 22 meeting between Anthropic and the White House. Any restoration, appeal or new compliance framework may depend on that process.
Source: Thorsten Meyer AI
