US Order Knocks Claude Fable 5 Offline Over Jailbreak

On Tuesday, Anthropic released Claude Fable 5. By Friday at 5:21pm ET, the company said a US government export control directive had forced it to take Fable 5 and Mythos 5 offline for all users.

The order cited national security concerns and focused on foreign national access by “any foreign national, whether inside or outside the United States, including foreign national Anthropic employees,” according to Wired. Anthropic said it removed access for everyone because that was the practical way to comply.

Friday 5:21pm directive forces Fable 5 and Mythos 5 offline

Anthropic said the government letter did not spell out the specific national security concern behind the directive. The company’s own reading is that officials believe someone found a way to bypass, or “jailbreak,” Fable 5.

“The letter did not provide specific details of its national security concern,” Anthropic wrote. “Our understanding is that the government believes it has become aware of a method of bypassing, or ‘jailbreaking’ Fable 5.”

Anthropic said it reviewed a demonstration of the technique. The company described the result as limited: a “small number of previously known, minor vulnerabilities,” which it said appeared “relatively simple.” It also said other publicly available models could discover them without the same bypass.

The shutdown hits both Claude Fable 5 and Mythos 5, even though the directive was framed around foreign national access. Anthropic said access to its other models “will not be affected,” but it has not provided a detailed customer transition plan in the supplied statement, a concern echoed by Microsoft’s Fable 5 data-risk block.

For product context on the model that was just released, see XOOMAR’s earlier coverage: Claude Fable 5 Sells Mythos-Class AI on a Short Leash.

Tuesday’s Fable 5 release becomes a federal safety test

The timing is what makes this sharp. Anthropic publicly released Claude Fable 5 earlier this week as a version of its Mythos model with safeguards blocking answers in areas including cybersecurity, biology, and chemistry. Before that, Mythos Preview had a limited rollout in April.

Anthropic said that earlier rollout was meant to let companies and organizations use Mythos’ cybersecurity capabilities defensively. The risk was obvious from the start: a model good at finding software flaws can help defenders, but the same capability can raise concern if guardrails fail.

A jailbreak claim matters because it suggests a user may be able to push a model past its safety rules. That can mean restricted outputs, behavior outside the developer’s intended controls, or instructions the model was designed not to provide. Anthropic argues the reported bypass does not show that kind of broad failure.

“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,” the company said.

Anthropic said no testers had found a “universal jailbreak,” meaning a broadly effective method for bypassing the model’s safeguards across a wide range of cyber capabilities. The company also said Fable had gone through “thousands of hours in total” of red-teaming with the US government, the UK AISI, third-party organizations, and internal teams.

The conflict now sits at the center of Anthropic’s public safety posture. Earlier this year, Trump’s Department of Defense labeled Anthropic a “supply chain risk” after the company sought limits on how the US military could use its technology, Wired reported. Anthropic then filed lawsuits against the Trump administration.

This is not a routine patch or a company-led rollback. It is a government-ordered suspension of access to commercial AI models. Anthropic says it supports a process for blocking unsafe deployments, but not this one.

“We believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts,” Anthropic wrote. “This action does not adhere to those principles.”

For readers tracking the AI security angle beyond this single shutdown, XOOMAR has also covered adjacent tooling and software-risk questions in 18B Artifacts Push Anthropic and JFrog Into AI Security.

Customers now wait on access, credits, and Anthropic’s promised 24-hour details

The immediate customer problem is simple: access to Fable 5 and Mythos 5 is being removed. The practical details are not.

Anthropic has not said in the supplied material when access will return, whether enterprise clients will get migration support, how API calls will fail, or whether customers will receive credits or refunds. The company apologized for the disruption and said it believes the directive reflects a misunderstanding.

Key unanswered operational questions now include:

• Access timing: When exactly will customers lose or regain access to Fable 5 and Mythos 5?
• Migration: Will Anthropic help customers shift workloads to other Claude models?
• Billing: Will affected customers receive credits, refunds, or contract adjustments?
• Technical review: Will Anthropic patch Fable 5, submit more evidence, or challenge the government’s assessment?
• Disclosure: Will the government identify the legal basis and technical trigger for the directive?

Anthropic said it would “share more details over the next 24 hours.” That is now the next hard reporting point.

The story changes if Anthropic produces a patch that satisfies the government, if officials disclose evidence of a broader jailbreak, if access is restored quickly, or if the dispute moves into court. It also changes if the same standard is applied to other frontier model releases, a scenario Anthropic explicitly warned about.

For now, Fable 5 is a test case. The question is not only whether one jailbreak worked. It is how fast US authorities can force an AI model offline when they decide its safeguards are not enough, and how much evidence they must show before doing it.

Impact Analysis

• The shutdown shows how quickly export controls can disrupt access to advanced AI models.
• Anthropic’s decision to remove access for all users highlights the compliance challenge created by foreign-national restrictions.
• The lack of disclosed details leaves customers uncertain about timelines, risk, and migration options.