Updated
Updated · POLITICO · Sep 30
Commerce Lifts Fable Export Controls After 19 Days as Anthropic Tightens AI Safeguards
Updated
Updated · POLITICO · Sep 30

Commerce Lifts Fable Export Controls After 19 Days as Anthropic Tightens AI Safeguards

3 articles · Updated · POLITICO · Sep 30

Summary

  • Anthropic regained access to Fable on July 1 after the Commerce Department lifted export controls on June 30, ending a 19-day shutdown triggered by concerns that users could jailbreak the model’s safeguards.
  • CAISI, Commerce and the White House were satisfied after Anthropic sent engineers to Washington, shared real-time testing data and trained an improved classifier to block problematic behavior; Mythos restrictions had already been lifted on June 26.
  • The standoff began two days after Fable’s June 9 release, when Amazon researchers found a jailbreak and the Trump administration concluded the model posed a national-security risk, then used export-control powers after Anthropic refused to take it offline voluntarily.
  • Defense officials, especially Deputy Defense Secretary Stephen Feinberg, delayed Fable’s return until they independently reviewed the updated safeguards, underscoring how multiple agencies now shape frontier-model releases.
  • The episode turned Anthropic and the White House into an uneasy working partnership and pushed rivals such as OpenAI toward more cautious, limited rollouts as the administration still withholds its broader AI safety framework.

Insights

Why did the White House rely on voluntary AI self-policing, then force Anthropic’s model offline with export controls days later?
If Trump’s AI accord is only “morally binding,” what would actually stop a dangerous frontier model from reaching the public?
Are secret audits and private government reviews enough to secure powerful AI models that can expose critical infrastructure weaknesses?