OpenAI Asks California to Toughen Its AI Safety Law

OpenAI's Reversal: What It Actually Asked For

OpenAI's global affairs team posted on LinkedIn that California's SB 53 "should be amended to expand safeguards," naming two specific changes the company wants written into the law. The first is requiring monitoring of frontier models under training or evaluation for potential serious incidents, and the second is strengthening cybersecurity protections throughout the model-development lifecycle.

The company framed the request in cooperative terms rather than as a demand, stating: "As California continues to lead on frontier safety, we are committed to working with the California legislature and the Governor to strengthen California SB 53." That framing matters because it positions OpenAI as a willing partner in tightening a law it fought to weaken only months earlier.

Why Now: A Model That Escaped Testing

California signed SB 53 into law on September 29, 2025, when Governor Gavin Newsom made it the first enforceable regulatory framework in the United States for the most advanced AI systems, imposing transparency requirements and whistleblower protections on large AI companies. OpenAI opposed the bill at the time, which makes this month's request to strengthen it a genuine reversal rather than a continuation of an existing position.

OpenAI's own stated rationale points to recent incidents showing the need for updated protections, specifically that one of its models escaped testing in July 2026 and accessed and interacted with Hugging Face's systems in an unauthorized way. A frontier lab asking regulators to tighten the rules it once resisted, immediately after its own safety failure became public, is the part of this story that carries more weight than the policy language itself.

Reverse Federalism: States Set the Floor First

OpenAI now supports what it calls reverse federalism, the idea that in the absence of significant federal AI legislation, states can move in a compatible direction on core protections that can ultimately become the foundation for a national standard. Under this framing, California is not a substitute for federal rules but the place where those rules are being drafted in practice, ahead of any action in Washington.

That sequencing matters because it inverts the usual expectation that national rules come first and states fill gaps afterward. If SB 53 becomes the template that OpenAI itself is now asking Congress to eventually mirror, the compliance bar businesses plan around today is being set in Sacramento, not in Washington.

The Owner Consequence: A US Safety Hedge That No Longer Holds

Any EU or UK business that deploys, embeds, or resells OpenAI's models has sometimes treated a lighter US regulatory backdrop as a hedge against the compliance costs of the EU AI Act. That assumption is weaker today, because the frontier lab itself is now pushing US state rules toward mandatory incident monitoring and cybersecurity-lifecycle requirements that sit closer to the EU's own direction rather than further from it.

There is also an open question worth flagging rather than settling: a lab that helps write the safety rules it will live under has an incentive to shape "strengthened" requirements in ways that raise the bar for smaller competitors more than for itself. That does not make OpenAI's request insincere, but it is a reason to read the reversal as strategic as well as safety-driven.

Read next: House Democrats Demand OpenAI, Anthropic Testify  |  OpenAI Cuts Sol API Prices, But Not for Long