OpenAI Urges California to Strengthen AI Safety Bill It Previously Opposed

In a letter to California legislators, OpenAI called for amendments to SB 53 that would impose stricter requirements on developers of the most powerful AI systems. The company previously opposed the bill, joining other tech firms in arguing that it could stifle innovation and drive AI development out of the state. Now, OpenAI says the bill should include stronger provisions for testing frontier models, greater transparency about capabilities and risks, and clearer liability for harm caused by AI systems that are deployed without adequate safeguards.

SB 53, introduced by state Senator Josh Becker, is one of several AI regulation efforts in California, home to many of the world's leading AI labs. The bill initially aimed to require safety testing for AI models trained on a certain threshold of computing power, but critics said its definitions were vague and its requirements too broad. OpenAI's reversal signals a dramatic shift in the political dynamics of AI regulation, as one of the largest and most visible AI companies now wants tighter rules—potentially setting a benchmark that other states and the federal government might follow.

Industry analysts say OpenAI's change of heart may reflect a desire to shape regulation on its own terms before more restrictive measures are imposed. The company has faced increasing scrutiny from regulators worldwide over the safety and societal impact of its products. By advocating for stronger provisions, OpenAI could also be attempting to differentiate itself from competitors like Anthropic and Meta, which have been more cautious about endorsing binding legislation.

However, not everyone is convinced that OpenAI's newfound support for tougher regulation is entirely altruistic. Some critics argue that the company's proposed amendments would favor its own business model—for example, by requiring costly safety audits that smaller startups would struggle to afford. Others worry that the bill could still be too weak or too narrowly focused, failing to address real-world harms like disinformation, bias, and labor displacement.

The bill now heads to committee for further revisions, and its fate remains uncertain. If passed, California could become the first U.S. state to enact comprehensive AI safety legislation, setting a precedent for the rest of the country and potentially influencing global regulatory standards.",

Why This Matters

OpenAI's policy reversal could reshape the debate on AI regulation in the U.S., potentially accelerating passage of SB 53 or inspiring similar efforts in other states and at the federal level.
The company's endorsement of tougher rules may signal that the industry sees regulation as inevitable and wants to help write the rules rather than face them from the outside.
For consumers, businesses, and policymakers, the outcome of SB 53 will set important precedents for how AI safety, transparency, and liability are handled in practice.

Background

California has emerged as the most active state-level battleground for AI regulation, with several bills targeting everything from deepfakes in elections to autonomous vehicles. SB 53, introduced in early 2025, specifically addresses "frontier" AI models—those with significant computational power and potential for harm. The bill mandates safety testing, incident reporting, and creation of an independent oversight body.

OpenAI initially opposed SB 53, arguing in prior statements that the bill's definitions were vague, its requirements overly burdensome, and that it could stifle innovation without effectively addressing safety. The company also warned that strict state-level laws could fragment the national market and push AI development to less regulated jurisdictions.

The reversal comes as OpenAI faces multiple investigations and lawsuits over its data practices, content moderation, and the safety of its models. Company executives have publicly called for more regulation in the past, but this is the first time it has actively advocated for a specific bill to be strengthened.

Key Perspectives

OpenAI: Wants SB 53 to include mandatory safety testing for frontier models, greater transparency about model capabilities and risks, and clearer liability rules. The company frames this as a necessary step to build public trust and ensure responsible development.

Other AI Developers (Anthropic, Meta, Google): Have expressed mixed reactions. Anthropic has been cautious about supporting binding legislation, preferring voluntary commitments. Meta and Google have not publicly commented on OpenAI's shift, but industry sources say they are concerned about the cost and scope of mandated testing.

Critics and Skeptics: Some consumer advocates argue OpenAI's proposals would entrench the dominance of well-funded companies by imposing compliance costs that startups cannot afford. Others say the bill still doesn't address key harms like algorithmic bias, misinformation, or job displacement. A few experts worry that strong state-level rules could create a patchwork of regulations that confuse the industry and undermine federal efforts.

What to Watch

Whether the California legislature's committees amend SB 53 to include OpenAI's specific proposals or adopt a different set of provisions.
The response from federal lawmakers—if California passes a strong AI safety bill, it could pressure Congress to act on a national framework.
How startups and smaller AI firms react: will they support or oppose the strengthened bill, given the potential for higher compliance costs?

ZOTPAPER