The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race
← Specific Events

Federal government

Scorecard

U.S. directive forces Anthropic to disable top-tier models after foreign-access restriction

What happened

Anthropic said it would abruptly disable its most advanced Fable 5 and Mythos 5 models for all users after a U.S. government directive suspended access for foreign nationals, based on a reported concern about a narrow potential jailbreak. Reuters reported that Anthropic said it had received only verbal evidence of the concern and warned that applying the same standard across the industry could effectively halt new frontier-model deployments. Axios later reported that President Trump had recently viewed Anthropic as a possible national-security threat, that Amazon had raised the vulnerability concern, and that the parties were working toward standards to evaluate AI jailbreaks.

Potential direction of power

Concentration · Concentration +1

The concentration-leaning element is that model access can be abruptly restricted through national-security and export-control channels that are not fully visible to users, researchers, firms, foreign-born developers, or downstream institutions. The practical authority over model release and access becomes concentrated among federal officials, frontier-model firms, cloud/distribution partners, and security evaluators.

This is a centralizing layer-control signal. The model layer, national-security review layer, cloud-distribution layer, export-control layer, and user-access layer become tightly coupled. A model capability concern can become an access-control decision affecting all users, with cloud providers and government agencies mediating practical availability.

Window of contestability

Narrows

Negative / mixed. The episode appears to move frontier-model access control into national-security, export-control, cloud-distribution, and company-by-company vulnerability channels. Some security review may be justified, but the public visibility, evidentiary standard, review mechanism, and appeal path remain unclear.

Supporting sources

  1. reuters.com · Supporting source 1 ↗
  2. axios.com · Supporting source 2 ↗