Anthropic adds new guardrails to Claude Fable 5 after Trump admin export controls

Anthropic has resolved an impasse with the Trump administration over its Claude Fable 5 AI model by agreeing to expand security guardrails. The model was previously subject to export controls after the administration learned users could circumvent restrictions by asking Fable 5 to fix code instead of identifying security issues—a technique highlighted in a paper by Amazon. The new safeguard extends an existing barrier to any request related to that specific behavior, automatically redirecting such queries to the less-capable Opus 4.8 model and notifying users the request was blocked.

Commerce Secretary Howard Lutnick confirmed the removal of restrictions in a letter, noting Anthropic committed to proactively detecting and addressing security risks. While the commerce department is satisfied, defense secretary Pete Hegseth still considers Anthropic a supply chain risk under a February 28 order, leaving the company's full reinstatement incomplete. The incident underscores the delicate relationship between frontier AI companies and national security regulations.