OpenAI pushes 'reverse federalism' as states align on AI safety

OpenAI’s Chief Global Affairs Officer, Chris Lehane, is making the case that a patchwork of state AI laws isn’t a bug—it’s a feature, at least temporarily. Writing in a new policy piece, Lehane argues that California, New York, and Illinois are creating a de facto national safety standard through what the company calls ‘reverse federalism.’ With Washington still working on a federal framework, these states are converging on three core pillars: mandatory safety frameworks with risk assessments, incident reporting, and independent audits.

California set the initial disclosure rules, New York proved the model could travel across state lines, and Illinois recently added a requirement for independent verification. Lehane is blunt about what he sees as the alternative—a chaotic ‘patchwork’ of laws that confuses consumers and bleeds developer resources, especially from startups. He warns that ‘performative or virtue signaling’ won’t cut it when the goal is a durable safety regime.

The strategy is clearly designed to put the U.S. in a position to export its democratic AI values globally. Lehane ties state-level alignment directly to national security, arguing that a coherent system is essential for getting advanced tools into the hands of government and critical infrastructure partners. His push for discipline is paired with a warning about mission creep: states shouldn’t be in the business of managing national security risks or conducting highly technical reviews better suited for federal experts with classified access.

Meanwhile, the Trump Administration is said to be working with technical experts on a framework for government testing of frontier models on cybersecurity. OpenAI confirms it’s in ‘constructive discussions’ with the administration and other stakeholders. The underlying message is unmistakable—the company wants the reins on safety kept tight, but held by a few aligned hands, not dozens of competing ones.