The company disputes the rationale behind the order, arguing that the reported vulnerabilities are limited and comparable to those in other AI models.
Anthropic has suspended global access to its Fable 5 and Mythos 5 AI models after receiving a US government export control directive citing national security concerns.
The company said the order requires it to halt access to both models for all foreign nationals, whether located inside or outside the United States, including Anthropic employees who are not US citizens. To comply, Anthropic has disabled Fable 5 and Mythos 5 for all customers worldwide. Access to its other AI models remains unaffected.
According to Anthropic, the directive was received at 5:21 p.m. ET and did not include detailed information about the underlying national security concerns. The company said it believes the government’s action is tied to a reported method of bypassing, or “jailbreaking,” Fable 5’s safeguards.
Anthropic said it reviewed a demonstration of the technique and found that it exposed only a small number of previously known, minor vulnerabilities. The company added that similar findings can be identified using other publicly available AI models without requiring any safeguard bypass.
The dispute centers on the effectiveness of Fable 5’s security protections. Anthropic said the model underwent thousands of hours of red-team testing before launch, involving the US government, the UK AI Safety Institute, third-party organizations, and internal teams.
According to the company, those tests showed Fable 5’s safeguards were more effective than those of earlier generations of AI models. Anthropic also said no tester had discovered a “universal jailbreak” capable of broadly circumventing the model’s protections across a wide range of cybersecurity-related tasks.
The company acknowledged that perfect jailbreak resistance remains unattainable across the industry. Instead, Anthropic said it adopted a “defense-in-depth” strategy designed to make jailbreaks either highly limited in scope or prohibitively expensive to develop.
As part of that approach, Anthropic introduced a 30-day data retention policy for Fable 5, allowing the company to monitor, investigate, and mitigate emerging jailbreak techniques. The policy has drawn scrutiny from some customers, but Anthropic argues that it is necessary to maintain model safety.
The company said it has not received evidence that any disclosed jailbreak resulted in harmful outcomes. Anthropic further claimed that the only evidence presented by the government involved a narrow technique that asked the model to review a software codebase and identify vulnerabilities.
After reviewing the report it believes informed the directive, Anthropic said the capabilities demonstrated were already widely available through other frontier AI models, including OpenAI’s GPT-5.5, and are routinely used by cybersecurity professionals.
Despite its objections, Anthropic said it is complying with the directive and has removed access to both Fable 5 and Mythos 5.
However, the company strongly disagreed with the decision, arguing that recalling a commercial AI model used by hundreds of millions of people based on a limited, non-universal jailbreak sets a problematic precedent.
Also Read: Karthik Ranganathan on Why AI’s Future Starts With Data Infrastructure
“If this standard were applied across the industry, it would effectively halt the deployment of new frontier AI models,” Anthropic said.
The company reiterated its support for government oversight of advanced AI systems but argued that such interventions should occur through a transparent, technically grounded, and clearly defined regulatory process.
Anthropic said it is working with authorities to resolve what it described as a misunderstanding and restore access to the models as quickly as possible.
