OpenAI's Chris Lehane outlines a 'reverse federalism' approach to US AI safety policy

OpenAI's Chief Global Affairs Officer Chris Lehane published a policy essay on July 15, 2026, arguing that state-level AI safety legislation is converging into a de facto national standard — a dynamic he calls "reverse federalism" — and that the federal government should build on that convergence rather than let a patchwork of state rules take hold.

What's new

The post credits California, New York, and Illinois with building the core of that emerging standard. In OpenAI's own words: "California, New York, and most recently Illinois have advanced frontier safety legislation that helps move the country toward a common baseline for governing the most powerful AI systems."

Lehane breaks down what each state contributed: "California established the core disclosure framework. New York showed the approach could be adopted across jurisdictions. Illinois complemented it by requiring independent verification of key disclosures." He distills the pattern into three elements OpenAI wants other states — and eventually the federal government — to converge on:

• A documented safety framework with risk assessments for frontier models and public disclosure of results
• Reporting requirements for serious safety incidents
• Governance and accountability through independent, objective audits

On the federal side, the post says the Trump Administration is "working with technical and national security experts on a framework for US government testing of the most capable AI models on cyber," and that OpenAI is "engaged in constructive discussions with the Administration, peer companies, business groups, and other stakeholders helping shape the effort." The piece also calls for strengthening the Center for AI Standards and Innovation (CAISI) as the federal body best positioned to run technical evaluations of frontier systems that individual states can't replicate.

Context

This follows a year in which state legislatures moved well ahead of Congress on frontier-model regulation, with California, New York, and Illinois each passing their own safety-disclosure laws in the absence of comprehensive federal legislation. OpenAI has published a string of related policy essays over the past year — including its frontier safety blueprint and public policy agenda — arguing consistently for federal primacy on frontier-model oversight while states focus on narrower, complementary rules.

The essay arrives as the White House has separately been shaping direct government-access arrangements with frontier labs, underscoring that federal AI policy is being negotiated on multiple tracks at once — testing standards for national-security uses on one side, state disclosure laws on the other.

Why it matters

OpenAI is making an explicit bid to shape which regulatory model wins: a small number of aligned state laws hardening into a de facto national standard, rather than 50 different state regimes or a slow-moving federal statute. The company's warning against "policy creep" and a fragmented patchwork is a direct argument to other state legislatures currently drafting their own AI bills — telling them, in effect, to copy California/New York/Illinois rather than write bespoke rules.

It also signals where OpenAI wants the center of gravity for the hardest oversight questions — cyber capability testing, national-security-relevant evaluations — to sit: with federal bodies like CAISI, not state regulators. For a company navigating safety disclosure obligations across a growing number of jurisdictions, getting states to converge on a shared framework would meaningfully simplify compliance while still keeping the toughest technical reviews in Washington's hands rather than a state capitol's.