Trump Administration's Fractured AI Policy Apparatus Targets Anthropic Over Export Controls
Key Takeaways
- ▸Trump administration's AI governance is fragmented across at least three competing power centers with incompatible regulatory philosophies, from moderate incentive-based approaches to hardline China containment
- ▸Anthropic faces concrete pressure through export controls and mandatory safeguard negotiations, with its Fable 5 model temporarily offline pending CAISI approval—serving as a test case for administration enforcement
- ▸National Cyber Director Cairncross is targeting 'distillation' practices (Chinese labs training models using outputs from US frontier models) as national security priority, while Commerce focuses on broader competitive advantage
Summary
As China's open-weight AI models grow more capable, the Trump administration has assembled competing power centers across multiple agencies to set AI export policy, but with little interagency coordination and fundamentally conflicting visions. Commerce Secretary Howard Lutnick, acting CAISI director Arvind Raman, and National Cyber Director Sean Cairncross represent three distinct regulatory approaches—ranging from Lutnick's incentive-based strategy to Cairncross's hardline restrictions on Chinese labs distilling US models. Anthropic has emerged as a central focal point of this bureaucratic tension, facing export controls from Lutnick's Commerce Department and weeks of intensive negotiations with CAISI staff over jailbreak safeguards for its Fable 5 model. With little centralized coordination, actual policy is likely to be determined by whichever official maintains the most influence with President Trump, creating unpredictability for AI companies navigating regulatory requirements and national security considerations.
- Lack of unified decision-making means individual relationships with President Trump will determine actual policy outcomes, not interagency consensus
