US orders Anthropic to suspend Fable 5 and Mythos 5 over security concerns

SAN FRANCISCO: In a dramatic twist for the AI industry, Anthropic has been forced to suspend access to its flagship models, Fable 5 and Mythos 5, after receiving an export control directive from the US government citing national security concerns.

The order, delivered at 5:21 pm ET on Thursday, requires Anthropic to block access to the two models for all foreign nationals, whether located inside or outside the United States. According to the company, complying with the directive effectively means disabling the models for all customers worldwide. Access to Anthropic’s other AI models remains unaffected.

The company said the government did not provide detailed information about the underlying national security concern. However, Anthropic believes the action is linked to an alleged “jailbreak” technique capable of bypassing some of the safeguards built into Fable 5.

After reviewing the reported method, the company said it identified only a small number of previously known vulnerabilities, describing them as minor and arguing that similar results can be achieved using other publicly available AI systems.

The dispute centres on the effectiveness of safety controls designed to prevent misuse of advanced AI models for cybersecurity-related tasks. Anthropic noted that before launching Fable 5, it conducted extensive safety testing involving government agencies, independent organisations and internal teams, accumulating thousands of hours of red-team evaluations.

According to the company, those tests found Fable’s safeguards to be stronger than those deployed in earlier generations of AI systems. Anthropic also stated that no researchers have yet demonstrated a universal jailbreak capable of broadly bypassing the model’s protections across a wide range of tasks.

The company acknowledged that perfect jailbreak resistance remains an unresolved challenge across the AI industry. Instead, it said it adopted a “defence in depth” approach designed to make successful attacks difficult, expensive and easier to detect through continuous monitoring and data retention measures.

Anthropic argued that the alleged vulnerability cited by authorities involved a narrow use case in which the model was asked to review software code and identify flaws. The company maintained that the capabilities demonstrated were comparable to those already available through competing AI systems, including GPT-5.5 from OpenAI.

The company warned that applying the same threshold across the industry could significantly slow or even halt the deployment of new frontier AI models.

While complying with the directive, Anthropic has publicly challenged the government’s decision-making process. The company said regulators should have the authority to block genuinely unsafe deployments, but argued that such interventions should be transparent, evidence-based and grounded in clear technical standards.

The move is likely to intensify debate around AI governance at a time when governments worldwide are seeking greater oversight of increasingly capable models. It also raises questions about how regulators should balance national security concerns with innovation and commercial deployment.

For now, millions of users are left waiting as Anthropic works to overturn what it describes as a misunderstanding. The company said it plans to release additional information within 24 hours and is seeking to restore access to Fable 5 and Mythos 5 as quickly as possible.