OpenAI Proposes 'Reverse Federalism' Vision for Unified State-and-Federal AI Governance

OpenAI’s Chief Global Affairs Officer Chris Lehane has published an article proposing an AI governance approach he calls “reverse federalism,” which combines state and federal law. Rather than waiting for Congress to act, he argues that leading states aligning on coordinated safety legislation can create a de facto national standard, while at the federal level, CAISI (the Center for AI Standards and Innovation) should be strengthened to align state and federal efforts.

Details

• What is reverse federalism: An approach in which major states adopt a shared legislative framework without waiting for federal law, effectively steering companies toward a national standard. California, New York, and Illinois have already moved ahead with similar legislation
• Three common pillars of state-level safety legislation:
  • Documenting and publishing safety frameworks, including risk assessments for frontier models
  • Mandatory reporting of serious safety incidents
  • Independent third-party audits and governance
• Recommendations for the federal level: Argues that CAISI should be strengthened for testing and evaluating frontier AI, with the federal government taking a leading role. The Trump administration is aiming to establish cybersecurity testing standards by early August, and OpenAI supports aligning both state and federal frameworks
• Concerns about regulatory fragmentation: Warns against “policy creep,” where states regulate independently and inconsistently, and against a patchwork of state laws that regulators struggle to enforce. Argues that areas touching national security should be handled by specialized federal agencies
• Vision for the future: Building on these domestic efforts, OpenAI aims to work with governments including the G7 to build a “U.S.-led international AI governance framework” grounded in democratic values

How to try it

This article is a policy proposal, not a product announcement. It doesn’t change how you use ChatGPT, but if you’re interested in the future direction of AI regulation, you can read the full text on OpenAI’s official site.