Anthropic has completely disabled access to two of its advanced language models following an emergency government order issued Friday evening. The company shut down Fable 5 and Mythos 5 to all users, including its own employees, after federal authorities cited unspecified national security concerns.

According to The Verge AI, the directive required the company to block access globally, affecting customers both within the United States and internationally. The order came without detailed written explanations of the specific threat, though government officials provided verbal briefings about the perceived vulnerability.

Vague Threat Assessment

In a public statement, Anthropic acknowledged compliance with the government mandate while expressing frustration about the lack of transparency. The company noted that authorities never formally documented their security concerns or provided written technical analysis. Instead, officials verbally described evidence of potential jailbreak attempts and claimed to have identified minor vulnerabilities that could be replicated through other available models.

This approach raises important questions about how the U.S. government evaluates risks from advanced AI systems and the standards it applies when restricting access to privately developed technology. The absence of formal documentation may complicate Anthropic's ability to understand, remediate, or publicly defend its position on the decision.

Industry and Policy Implications

The incident reflects growing government interest in managing the development and deployment of frontier AI models. Federal agencies have increasingly asserted authority over large language model releases, though the legal and regulatory framework governing such interventions remains unsettled. The emergency nature of Friday's order suggests escalating concerns about potential misuse or capabilities that officials view as genuinely hazardous.

Key developments from this action include:

  • Immediate operational disruption for Anthropic customers relying on Fable 5 and Mythos 5
  • Establishment of government precedent for unilateral model access restriction
  • Questions about the proportionality of shutting down access entirely versus implementing targeted restrictions
  • Lack of formal documentation making it difficult for external experts to assess government threat claims

The situation underscores tension between rapid AI development, national security interests, and the principle of public accountability. Tech companies have generally accepted some level of government oversight in sensitive areas, but the absence of clear criteria or written justification complicates both industry compliance and public understanding of how these decisions are made.

Anthropic's compliance demonstrates the substantial power federal authorities now wield over private AI development, even when the specific technical justification remains opaque. Whether this approach becomes standard practice could reshape how companies develop, test, and release advanced language models in the coming years.