TL;DR
The Commerce Department placed Anthropic’s Claude Fable 5 and Mythos 5 under export controls on June 12, forcing the company to disable both models worldwide. The government’s stated rationale remains contested, but the shutdown has made regulatory interruption a concrete risk for enterprise AI buyers.
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 worldwide and raising a new reliability question for businesses building on frontier AI systems.
According to Anthropic’s account, Commerce Secretary Howard Lutnick sent CEO Dario Amodei a letter at 5:21 p.m. ET placing the models under export controls and barring access by foreign nationals, including foreign-national employees inside Anthropic. The company said it had no clean way to comply selectively and turned off both models for all customers by midnight.
Claude Fable 5 had launched three days earlier as the public, heavily guarded version of Anthropic’s new system. Mythos 5, described in the source material as the more powerful underlying model, was routed to selected organizations for cyber-defense work under Project Glasswing rather than released openly.
The government has not publicly released a full technical explanation for the order. Anthropic said it understood the action to be tied to a reported jailbreak of Fable 5 and called the directive a “misunderstanding.” Reports cited in the source material from The Wall Street Journal, Semafor and Axios point to concerns about malicious cyber use, possible model access by a China-linked group and reverse-engineering risk.
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.
Shutdown Risk Hits AI Buyers
The immediate cost is not only that two models went offline. The larger impact is that customers now have a live example of a frontier AI service being cut off by government order on short notice, even for users outside the dispute.
That matters for banks, cloud providers, software companies, defense contractors and health-science users that are weighing whether to build core workflows around proprietary frontier models. If access can be interrupted globally, buyers may add export-control risk to pricing, procurement and continuity planning.
The episode may also strengthen demand for multi-model systems and self-hosted open-weight models. Those options can carry their own security, support and performance limits, but they are harder for a single vendor or regulator to revoke once deployed on a customer’s own infrastructure.
AI Engineering: Building Applications with Foundation Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Three Days From Launch
Anthropic released the Mythos-class models on June 9, presenting them as advanced systems for cybersecurity and biomedical work. By June 12, the models had shifted from product launch to national-security dispute.
The source material says U.K. AI Safety Institute red-team work found a jailbreak that produced malicious answers and later multi-step agentic tool use. The Wall Street Journal reported that Amazon researchers also raised alarms, and that Amazon CEO Andy Jassy contacted Treasury Secretary Scott Bessent about cyberattack-usable information obtained from Fable 5.
Anthropic and more than 120 cybersecurity executives and engineers have pushed back, arguing in an open letter that the models’ capabilities are not unique and that similar security tasks can be performed by other U.S. and Chinese systems. Their position is that restrictions may remove tools from defenders while attackers keep access to other models.
“misunderstanding”
— Anthropic
Applied AI Governance: The Model Context Protocol as an Enterprise Control Plane for Autonomous Agents
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Jailbreak Details Remain Disputed
It is not yet clear how severe the reported jailbreak was, whether it gave access to capabilities unavailable in peer models, or whether a China-linked group obtained the model. The government has not publicly laid out the full evidence behind the order.
It is also unresolved whether export controls designed for physical goods and chips can be applied cleanly to software models used by global customers and multinational workforces. Anthropic says the order left it unable to comply without disabling access for everyone.
Combating Cyberattacks Targeting the AI Ecosystem: Assessing Threats, Risks, and Vulnerabilities
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
White House Talks Loom
Anthropic and White House officials are expected to meet on June 22. The key question is whether the administration narrows, lifts or defends the controls, and whether it publishes a clearer standard for when a model must be pulled.
AI buyers will be watching for service-restoration terms, disclosure from Commerce and any signal that other frontier models could face similar restrictions. Until then, the shutdown is likely to accelerate backup planning across enterprise AI deployments.
Operating AI: Bridging the Gap Between Technology and Business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What happened to Claude Fable 5 and Mythos 5?
The Commerce Department placed both models under export controls on June 12. Anthropic said the order barred foreign-national access and forced it to disable both models globally.
Why did the U.S. government act?
The full rationale has not been made public. Reports cited in the source material point to concerns about jailbreaks, cyberattack-usable outputs, possible access by a China-linked group and reverse-engineering risk.
Does this mean the models were proven unsafe?
No public record in the source material proves that. Anthropic says the issue was narrow and not a universal jailbreak. Government-linked and third-party accounts claim the risks were serious enough to warrant controls.
Why does this matter for companies using AI?
The shutdown shows that access to a proprietary frontier model can be interrupted by government action. That may push companies toward backup vendors, multi-model designs and self-hosted systems.
When could access return?
No restart date has been confirmed. Anthropic and White House officials are expected to meet on June 22, when the scope and future of the controls may become clearer.
Source: Thorsten Meyer AI
