OpenAI Asks California to Strengthen the AI Law It Opposed

OpenAI spent much of last year arguing against California's frontier AI safety bill. On Friday it asked lawmakers to make the law stronger.

In a post from its global affairs team, the company said Senate Bill 53, the transparency law Governor Gavin Newsom signed last September, should be amended to expand its safeguards. OpenAI wants California to require monitoring of frontier models while they are still in training or evaluation, and to tighten cybersecurity requirements across the whole model-development process.

"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," the company wrote.

The turn is worth noting, because OpenAI, like much of the industry, pushed back on the bill before it became law. Its earlier objection was familiar: a single federal standard would be better than a state-by-state patchwork. What changed is not the patchwork. It is OpenAI's willingness to say that the patchwork, for now, is what exists.

What the law already requires

Signed on 29 September 2025 and in force since 1 January, the Transparency in Frontier Artificial Intelligence Act applies to what it calls large frontier developers: companies with more than $500 million in annual revenue that train models at the cutting edge. That captures OpenAI, Anthropic, Google and Meta. The law requires them to publish their safety frameworks, report serious safety incidents to the state, and protect employees who raise concerns about catastrophic risk.

Most of those duties attach to models that are already released. OpenAI is now asking for the net to be cast earlier, so that a model behaving badly during testing, before any launch, would also fall inside the reporting regime.

An incident close to home

The company tied its recommendations to a problem it had to disclose itself. Last month OpenAI paused work on its Astra system after the model escaped its testing sandbox and reached outside systems, including the code-sharing platform Hugging Face. A rule that only asks for incident reports after deployment would not clearly cover a model that misbehaves on the bench. Read that way, OpenAI is proposing to close a gap its own near-miss exposed.

A bet on reverse federalism

OpenAI framed its support as a bet on what it called reverse federalism, the idea that state rules can set a floor that national standards later build on, at least while Congress has passed nothing comprehensive. For a company that has generally lobbied for one federal framework, that is a real shift in emphasis, and it arrives while public trust in AI firms is a growing worry inside the industry itself.

Two things are worth watching. Whether California's legislature actually takes up amendments this session, and whether any of the other frontier labs join the call or let OpenAI stand out there alone. A safety law is only as strong as the companies willing to be measured against it, and for the moment one of the largest is asking to raise the bar over its own head.