Highlights

The U.S. government issued an emergency export-control directive requiring Anthropic to suspend access to its two newest models, Claude Fable 5 and Claude Mythos 5, over a reported national security concern about a possible jailbreak. Anthropic complied but disputed the government’s assessment, saying the reported vulnerability appears simple and is already reproducible on other public models like GPT-5.5. The company warned the order could set a dangerous industry-wide precedent that would slow future frontier-model rollouts. This action restricts foreign access to the models and temporarily disabled them for all customers to ensure compliance.

Sentiment Analysis

• Overall sentiment in this report is mixed-to-cautionary. The government’s move signals serious national-security concern and urgency, which is negative for Anthropic and potentially for broader model deployments. At the same time, Anthropic’s rebuttal and characterization of the issue as a relatively simple, already-known vulnerability introduce skepticism about the necessity and proportionality of the restriction. Stakeholders are divided: regulators emphasize risk and control; the company emphasizes technical context and precedent risk for the industry.


• 
  
  
  

  
  60%
  

  
  
  
  

Article Text

The U.S. government issued an emergency export-control directive that ordered Anthropic to immediately suspend access to its two most advanced artificial intelligence models, Claude Fable 5 and Claude Mythos 5. The directive, released shortly after the models’ commercial debut, bars any foreign national from accessing them, whether located inside or outside the United States. To comply, Anthropic disabled the models for its entire customer base, a step the company said was necessary to ensure full adherence to the order.

The government has cited a national security concern tied to a reported method for bypassing the models’ safeguards — a so-called "jailbreak." Although the public notice did not lay out technical specifics, officials believe a demonstration revealed a technique that could circumvent protections in the Fable 5 model. Mythos 5, which was only available to a small set of partners and has fewer guardrails, was singled out because of its heightened capability to identify cybersecurity exploits.

Anthropic disputed the gravity of the finding, saying it reviewed the demonstration and judged the vulnerability to be fairly simple. The company argued that the identified behavior is already replicable with other publicly available models, including OpenAI’s GPT-5.5, and therefore not unique to its releases. Anthropic also said that so far the government has provided only verbal evidence of what it described as a narrow, non-universal jailbreak — essentially asking a model to read a specific codebase and propose fixes for software flaws.

While complying with the directive, Anthropic warned that the government’s action could set a damaging precedent. The firm said that if the same standard were applied across the industry, it could effectively halt deployments of new frontier models by all major providers. The company emphasized the broader implications for innovation and deployment timelines while noting that access to its other models remains unaffected and that it is working to restore service as soon as possible.

The public discourse around the event quickly grew heated. David Sacks, co-chair of the President’s Council of Advisers on Science and Technology, posted that a trusted partner who tested Fable had found a jailbreak and that the administration asked Anthropic to fix it or remove the model from deployment. According to Sacks, Anthropic declined to remove the model, prompting the administration to issue the export-control order. He characterized the administration’s action as reluctant but necessary and said officials believe the issue should be fixable by Anthropic.

The episode follows earlier tensions between Anthropic and U.S. authorities, when the company resisted certain government requests related to expanded surveillance agreements and autonomous weaponry. Those disagreements led to public rebukes and a Department of Defense designation of Anthropic as a potential supply-chain risk, a label the company has contested in court. Despite earlier friction, there were subsequent reports of renewed government interest in Anthropic’s models for certain use cases, highlighting the complicated relationship between national-security needs and commercial AI development.

Observers note several broader themes. First, the incident underscores how quickly regulatory and national-security considerations can affect commercial AI operations, particularly for models deemed "frontier" in capability. Second, it highlights the challenge of communicating technical risk assessments between private developers and government agencies: what one side regards as a manageable, well-understood vulnerability may be viewed by authorities as an unacceptable exposure. This disconnect over risk assessment and disclosure can produce abrupt policy actions with major commercial consequences.

Finally, the directive raises questions about international access controls and the balance between openness in AI research and protecting against potential misuse. If export-control style restrictions become a common tool for managing AI risk, companies and governments will need clearer frameworks for rapid evaluation, technical validation, and remediation so that innovation is not unduly stalled while genuine vulnerabilities are addressed.

Key Insights Table