Anthropic Takes Claude Fable 5 Offline Following US Government Order

Anthropic Shuts Down Claude Fable 5 Amid Government Pressure

In an extraordinary and unprecedented move, Anthropic has taken its Claude Fable 5 AI model completely offline following a directive from the US government — marking yet another major point of conflict between the AI company and the Trump administration.

Although the government's order specifically targeted access for foreign nationals — including non-US citizens employed by Anthropic itself, regardless of their physical location — the company chose to suspend access for its entire customer base in order to ensure full compliance with the directive.

A Rocky Relationship With the Trump Administration

This latest development is part of a broader and increasingly contentious relationship between Anthropic and federal authorities. Earlier this year, the Trump administration's Department of Defense formally classified Anthropic as a "supply chain risk." That designation came after the company attempted to establish boundaries around how the US military could deploy its artificial intelligence technology. The classification effectively blocked government agencies and their contractors from utilizing any of Anthropic's products or services. In response, Anthropic took legal action and filed lawsuits directly against the Trump administration.

What Is Claude Fable 5?

Claude Fable 5 is a specialized version of Anthropic's Mythos AI model, built with enhanced safeguards that restrict it from responding to queries related to cybersecurity, biology, and chemistry. The model was officially made available to the public on a Tuesday, following what Anthropic described as a collaborative process with the US government.

Prior to its full public launch, a limited version known as the Mythos Preview had been rolled out in April. That early release was designed to allow select companies and organizations to leverage the model's advanced cybersecurity capabilities to strengthen their own defenses — while also addressing concerns that such powerful technology could potentially be weaponized by malicious actors to develop sophisticated hacking tools.

Government Cites Jailbreak Concerns

According to a blog post published by Anthropic on Friday, the company received a letter from the US government at 5:21 PM ET. Notably, the letter offered no specific details regarding the nature of the national security concern that prompted the order.

However, Anthropic indicated that the government's concern centers on a believed method of bypassing the model's built-in safety restrictions — commonly referred to as "jailbreaking." In the blog post, the company explained its understanding of the situation:

"Our understanding is that the government believes it has become aware of a method of bypassing, or 'jailbreaking' Fable 5."

Anthropic pushed back on the severity of this claim, stating that its own review of the demonstrated technique revealed only a small number of previously identified, minor vulnerabilities. The company also noted that these same vulnerabilities could be uncovered using other publicly available AI models — without any need for a bypass.

Anthropic Disputes the Threat Level

In its public statement, Anthropic argued strongly that Claude Fable 5 is protected by robust safeguards designed to minimize the risk of misuse. The company characterized the alleged jailbreak as narrow in scope and unlikely to provide any attacker with a meaningful advantage over what they could already achieve using other AI tools currently available on the market.

"To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws," Anthropic stated in its blog post.

Requests for comment directed to spokespeople for both the White House and the US Commerce Department went unanswered at the time of publication.

Amodei Calls for Transparent AI Oversight

Earlier in the week, Anthropic CEO Dario Amodei had published a policy essay outlining his support for a fair, structured, and transparent governmental process for evaluating and — when genuinely necessary — blocking the release of AI models deemed unsafe. However, in Friday's blog post, the company made clear that it does not view the current government action as meeting that standard, stating plainly that "this action does not adhere to those principles."

The situation continues to evolve, raising significant questions about the future relationship between private AI developers and US federal authorities — and how national security concerns will shape the development and deployment of advanced artificial intelligence.