Anthropic Pulls Two New AI Models Offline After Government Order
Anthropic has suspended customer access to its newly launched Fable 5 and Mythos 5 models after receiving a US government order on Friday, June 12, according to the company’s public explanation cited by Engadget. The move affects all customers of those two models, while the rest of Anthropic’s model lineup and its Claude chatbot remain available.
The company said the directive required it to halt access for foreign nationals, whether they are inside or outside the United States and even if they work at Anthropic. Anthropic linked the intervention to national security concerns, though it also said the government did not spell out those concerns in detail.
That leaves an unusual and significant development in the fast-moving AI market: a major commercial model launch interrupted within days by a national security order, with the company complying while also signaling disagreement with the basis for the recall.
What Anthropic Said Prompted the Action
Anthropic said it believes the government acted after hearing about a method of jailbreaking Fable 5. In the company’s account, officials provided verbal evidence about one potential narrow, non-universal jailbreak shared by an unnamed entity. Anthropic said it would provide more detail within 24 hours.
The distinction matters. Anthropic argued that no model provider can guarantee perfect resistance to jailbreak attempts and that all models are vulnerable to attacks designed specifically for them. Its stated strategy for Fable 5 was to make jailbreaks either narrow or expensive to produce, while relying on monitoring to detect and shut down successful attempts.
That framing suggests the company sees the issue less as proof of a uniquely dangerous failure and more as a known security problem that must be managed through layered defenses. Even so, the immediate practical result was a full suspension of access to the affected models.
Why Fable 5 and Mythos 5 Matter
The order lands at a sensitive moment because Anthropic had only just launched Fable on June 9. The company positioned Fable as a way to bring many of Mythos’ capabilities to a wider public audience. Mythos, by contrast, had been reserved for Project Glasswing partners and described as Anthropic’s state-of-the-art cybersecurity model.
Anthropic also said Fable’s capabilities exceed those of any previous model it has released. In company testing referenced by Engadget, Fable beat Pokemon FireRed, while Claude had failed to beat the earlier Pokemon Red. That comparison is lighthearted on the surface, but it was presented as evidence that Fable represented a notable step up in capability.
In other words, this was not a marginal product update. It was a fresh flagship-class release tied to some of Anthropic’s most advanced work. Pulling it so soon after launch raises questions not only about security review, but also about how governments may intervene when frontier systems are perceived to create new risks.
A Test Case for AI Governance
Anthropic has been one of the more vocal large AI companies warning about the need for stronger oversight. Its response to the order reflects that position, but with an important caveat. The company said it believes the government should have the power to block unsafe deployments, yet only through a statutory process that is transparent, fair, clear and grounded in technical facts.
That is a notable public stance because Anthropic is simultaneously complying with the order and challenging the way it was carried out. Based on the source text, the company’s objection is not to the idea of oversight itself. It is to a recall process triggered by verbal evidence of a potential jailbreak, without the kind of clear procedural framework Anthropic says should govern such decisions.
The episode therefore goes beyond one company and one model family. It highlights an unresolved policy question for advanced AI systems: when does a reported exploit justify emergency intervention, and what standard of evidence should be required before governments can force a shutdown?
What Comes Next
For now, Anthropic customers are left with a narrow but important disruption. Fable 5 and Mythos 5 are unavailable, but other models and Claude are not affected. That limits the operational blast radius, yet the symbolism is large. A government order has reached directly into the deployment of a newly launched frontier model, and a leading AI developer has had to reverse access almost immediately.
Several unknowns remain based on the available reporting:
- What specific national security concern officials identified.
- How broadly the alleged jailbreak worked in practice.
- Whether the suspension is temporary or could turn into a longer rollback.
- What future disclosure Anthropic will provide about the exploit and its safeguards.
Those details will determine whether this is remembered as a precautionary pause or as an early template for direct state intervention in commercial AI releases. Either way, the message to the industry is already clear: frontier model launches are no longer governed only by technical readiness and market timing. They can also be interrupted by national security authorities in real time.
That shifts the center of gravity for AI deployment. Providers may still compete on performance and safety, but they are now operating in an environment where a reported vulnerability can trigger not just internal mitigation, but external enforcement. Anthropic’s suspension of Fable 5 and Mythos 5 is an early sign that the next phase of AI competition will be shaped as much by governance mechanisms as by model capability itself.
This article is based on reporting by Engadget. Read the original article.
Originally published on engadget.com
