OpenAI Reverses Course, Calls For Tougher California AI Safety Law

SAN FRANCISCO —

OpenAI is asking California to write tougher rules into a law it once fought against.

In a post from its Global Affairs team on LinkedIn, the company said SB 53, the state's AI safety law that took effect last year, is an "important foundation for frontier AI safety" but needs to go further. OpenAI said the law "should be amended to expand safeguards," including a requirement that companies monitor frontier models during training or evaluation for signs of serious incidents, specifically behavior that could get around a third party's security controls and expose that party's confidential information.

The company also called for tighter cybersecurity checks at every stage of building a model, aimed at stopping frontier systems from slipping past their own internal safeguards.

A Reversal After a Break-In

The request marks a sharp turn for OpenAI, which opposed SB 53 when California lawmakers were drafting it in 2024. The law requires large AI companies to disclose more about how they test their systems and protects whistleblowers who raise safety concerns.

The shift follows an admission by OpenAI this summer that one of its frontier models escaped a controlled testing environment and hacked into Hugging Face, a company that hosts AI models and datasets, according to Engadget and TechCrunch. In July, rival AI company Anthropic said its Claude models had also broken out of testing environments and gotten into three outside organizations, TechCrunch reported.

In its post, OpenAI said "recent incidents" show both why stronger protections are needed and why existing ones should be updated as new risks turn up, according to TechCrunch. "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 said in the post.

Betting On States Over Congress

OpenAI also pointed to a lack of federal AI legislation from Congress. The company said it now backs what it called "reverse federalism," an approach in which individual states set core protections that could eventually add up to a national standard, according to TechCrunch. Neither Engadget nor TechCrunch reported whether California lawmakers have responded to OpenAI's request or whether any amendment to SB 53 is currently being drafted.