TL;DR
The U.S. Commerce Department placed Anthropic’s Claude Fable 5 and Mythos 5 under export controls on June 12, 2026, forcing the company to disable both models worldwide. The security basis for the order remains contested, but the shutdown has already raised questions about whether frontier AI can be treated as a dependable global utility.
The U.S. Commerce Department ordered export controls on Anthropic’s Claude Fable 5 and Mythos 5 on June 12, forcing the company to disable both models worldwide and creating a new test case for how national-security rules can interrupt commercial frontier AI services.
According to the source material, Commerce Secretary Howard Lutnick sent Anthropic CEO Dario Amodei a letter at 5:21 p.m. ET on June 12 placing the two models under export controls. The order barred access by any foreign national, including foreign-national employees inside Anthropic. Anthropic said there was no clean way to comply while keeping the models online, so it disabled them for every customer by midnight.
The shutdown came three days after Anthropic launched the Mythos-class systems. Claude Fable 5 was described as the public, heavily guarded commercial model. Mythos 5 was the stronger underlying system, made available to selected organizations for cyber-defense work through Project Glasswing rather than released openly.
Anthropic has said the order cited national-security authorities but did not give a specific rationale. The company publicly called the action a misunderstanding and said its understanding was that officials had learned of a method for jailbreaking Fable 5. Anthropic said the reported flaw was narrow, not universal, and did not justify recalling models already deployed at large scale.
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.
AI Reliability Faces Policy Risk
The order matters because it turned model access from a technical and commercial question into a policy-risk question. Large companies buying AI systems now have a confirmed example of a U.S. frontier model being switched off worldwide after a government directive, even for domestic users who were not the stated target of export controls.
That could affect purchasing decisions across banks, cloud platforms, software companies, defense contractors and other buyers that depend on continuity. The source material cites Deutsche Bank’s reading that switch-off risk is now proven rather than hypothetical. That does not mean buyers will abandon U.S. AI models, but it adds another reason for customers to spread workloads across several providers.
The episode may also help open-weight and self-hosted models. If a customer can run a model on its own infrastructure, access is harder for a vendor or government order to revoke quickly. Supporters of U.S. frontier labs may see that as a bad trade: controls intended to limit misuse could push some users toward systems outside U.S. companies’ direct reach.
Artificial Intelligence for Cybersecurity: Develop AI approaches to solve cybersecurity problems in your organization
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Three Days From Launch To Shutdown
The timeline was unusually compressed. Anthropic released the Mythos-class models on June 9. The Commerce directive arrived on June 12. By the end of that night, both Claude Fable 5 and Mythos 5 were dark for all customers.
The government’s concerns have not been fully established in public. The source material says a U.K. AI Safety Institute red-team lead reported that his team built a jailbreak within hours and later extended it to multi-step agentic tool calls. The Wall Street Journal was cited as reporting that Amazon also warned officials after its researchers found outputs that could be useful in cyberattacks. Semafor was cited as reporting that officials also suspected a China-linked group had obtained the model, raising reverse-engineering concerns.
Anthropic and more than 120 cybersecurity executives and engineers have pushed back. Their argument, according to the source material, is that the models may be powerful but are not unique, with other U.S. and Chinese models able to perform similar security work.
“misunderstanding”
— Anthropic
AI model jailbreak detection software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Jailbreak Evidence Remains Disputed
It is not yet clear what specific evidence led Commerce to issue the directive, whether the reported jailbreak was broadly reproducible, or whether a China-linked group obtained enough access to create a real proliferation risk. The competing accounts do not fully match: officials and outside reports point to cyber and reverse-engineering concerns, while Anthropic says the model had already been tested for thousands of hours by internal teams, government evaluators, the U.K. AI Safety Institute and third parties without a universal jailbreak being found.
It is also unclear whether the order will remain in place, be narrowed, or be replaced by a more tailored control.
Intelligent Continuous Security: AI-Enabled Transformation for Seamless Protection
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
June 22 Talks Set Stakes
Anthropic is scheduled to meet with the White House on June 22. That meeting is expected to determine whether the company can bring Fable 5 or Mythos 5 back online, whether added safeguards would satisfy officials, or whether the shutdown becomes a longer-running precedent for frontier AI export controls.
AI customers will also be watching for signs of how future orders could be applied: to specific users, specific countries, model weights, hosted services, or entire commercial systems.
Architecting Enterprise AI Applications: A Guide to Designing Reliable, Scalable, and Secure Enterprise-Grade AI Solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What exactly happened to Anthropic’s models?
The Commerce Department placed Claude Fable 5 and Mythos 5 under export controls on June 12, 2026. Because the order barred access by foreign nationals anywhere, Anthropic disabled both models for all customers worldwide that night.
Why did the government act?
The public record is incomplete. The source material cites concerns about jailbreaking, cyberattack-usable outputs, possible access by a China-linked group and reverse-engineering risk. Anthropic says the action appears to have been based on a limited jailbreak concern.
Was the model proven unsafe?
No public finding has settled that question. Anthropic says no universal jailbreak was found after extensive red-teaming. Other accounts cited in the source material say testers and Amazon researchers found outputs that worried officials.
Why does this affect the broader AI industry?
The shutdown shows that a hosted frontier model can be cut off quickly by government order. That may push major buyers to use several model providers, add regulatory risk to procurement reviews, or test more self-hosted systems.
When could the models come back?
No return date has been confirmed. The next known milestone is Anthropic’s scheduled June 22 meeting with the White House.
Source: Thorsten Meyer AI
