US Government Forces Anthropic to Shut Down Advanced AI Models Over Security Fears

Government Intervention in Frontier AI Development

Anthropic has temporarily removed access to its most capable AI models, Fable 5 and Mythos 5, after receiving a direct export control directive from the U.S. government. The order, issued late on Friday, cited national security risks tied to a reported technique that could bypass certain safeguards in the new systems.

According to the company, the directive from the Department of Commerce required blocking access for all foreign nationals, including those working inside the United States. Unable to implement nationality-based restrictions reliably, Anthropic chose to suspend the models entirely for every user. Other Claude models, such as Opus 4.8, remain fully operational and unaffected.

Details of the Reported Vulnerability

The government's concern centers on a narrow method described as a potential jailbreak. It reportedly involves prompting the model to analyze specific codebases for vulnerabilities. Anthropic maintains that this technique is limited in scope and does not represent a universal bypass of the model's safety systems.

The company noted that similar vulnerabilities can be identified by other publicly available models without any special prompting. It emphasized that extensive red-teaming involving thousands of hours had been conducted prior to launch in collaboration with U.S. government entities, the UK AI Safety Institute, and independent organizations.

Anthropic believes the directive may stem from a misunderstanding and is actively working with officials to clarify the situation and restore access as soon as possible.

Context of the Models' Launch

Fable 5 was positioned as a significant advancement, capable of handling longer and more complex tasks than previous versions. It showed strong results in areas including software engineering, scientific research, finance, vision capabilities, memory, and general knowledge work.

However, the launch faced immediate scrutiny. Users reported overly restrictive filters that routed certain technical queries to older models. Additional concerns arose regarding data retention policies and transparency around hidden safeguards in the system card. Anthropic quickly adjusted some policies in response to feedback.

Broader Implications for AI Industry

This incident marks a notable example of direct government involvement in the rollout of frontier AI technology. It highlights ongoing tensions between rapid commercial development and national security considerations in the AI sector.

Anthropic continues its infrastructure expansion plans amid the restrictions. The company is pursuing significant financing for additional computing resources while also advocating for clearer regulatory frameworks around high-risk AI deployments, independent evaluations, and cybersecurity measures.

The situation occurs as Anthropic prepares for future growth, including reported steps toward a potential public offering. Industry observers note that such interventions could influence how other leading AI developers approach releases of advanced systems going forward.

Anthropic has indicated it will provide additional updates within 24 hours as discussions with authorities progress. The resolution timeline remains unclear, with some officials suggesting security reviews could extend for several weeks.