Anthropic said Thursday it is disabling access to its Fable 5 and Mythos 5 AI models for all customers after receiving a U.S. government directive tied to national security concerns. The company said the order requires it to block access by any foreign national, including foreign national employees, whether they are inside or outside the United States.
The effect, Anthropic said, is a broad shutdown of the two models rather than a limited restriction on specific users. The company said access to its other models will remain unchanged.
Anthropic said it received the directive at 5:21 p.m. Eastern time and that the government did not provide detailed public explanation of the concern. According to the company, officials believe they have identified a way to bypass or “jailbreak” Fable 5. Anthropic said it reviewed a demonstration of the technique and found that it exposed only a small set of previously known, minor vulnerabilities.
In its statement, Anthropic argued that the vulnerabilities identified in the report appear relatively simple and are not unique to its models. The company said similar issues can also be uncovered by other publicly available systems without any special bypass method.
Anthropic said it has not been shown evidence of a harmful jailbreak that would broadly undermine the model’s safeguards. The company described the issue as a narrow, non-universal jailbreak, rather than a wider attack capable of unlocking a broad range of cyber capabilities.
The company also said it believes the government’s action is too severe for the finding at hand. Anthropic said if the same standard were applied across the AI industry, it could effectively stop new model launches by frontier AI providers.
Anthropic defended the safety measures it had built into Fable 5, saying the model was launched with strong protections intended to reduce misuse in cybersecurity and other sensitive areas. The company said those safeguards had been tested for thousands of hours with help from U.S. and U.K. government bodies, third-party organizations, and internal teams.
According to Anthropic, those evaluations found Fable 5’s protections to be stronger than any previously deployed model. The company said no tester had yet found a universal jailbreak that could broadly bypass its defenses.
Still, Anthropic said it has never assumed perfect jailbreak resistance is possible. The company said it uses a defense-in-depth approach designed to keep potential bypasses narrow or expensive to carry out, while also monitoring for attacks and trying to shut them down quickly.
Anthropic said that approach also led it to require 30-day retention of customer data for Fable, a policy it described as costly but important for research into jailbreaks and mitigation efforts.
Anthropic said it is following the government’s legal directive and removing access to Fable 5 and Mythos 5 for all users, but it disagrees with the order and wants the restriction reversed. The company said it is working to restore access as soon as possible.
The company also said it plans to provide more details within 24 hours. Anthropic apologized to customers for the disruption and said it believes the situation reflects a misunderstanding.
The dispute comes as governments around the world are scrutinizing advanced AI systems more closely, especially over cybersecurity and national security risks. Anthropic said it supports government authority to block unsafe deployments, but only through a transparent and technically grounded process. In this case, it said the directive does not meet that standard.