California’s landmark AI transparency bill is officially law

Last year, California Governor Gavin Newsom killed the controversial SB 1047 bill because it was too harsh on AI innovation. Now, its softer successor just crossed the finish line.

Newsom officially signed Senate Bill 53, also known as the Transparency in Frontier Artificial Intelligence Act, making California the first U.S. state to legally mandate safety transparency from frontier AI developers.

Transparency without the liability trap

Authored by State Senator Scott Wiener, SB 53 was born out of a 52-page researcher report commissioned by Newsom after his SB 1047 veto. The original bill targeted models costing over $100 million to train with mandatory pre-deployment testing. SB 53 takes a much gentler path focused on disclosures rather than rigid pre-training bottlenecks.

Under the new law, frontier AI developers must publicly publish a framework detailing how their safety and security protocols incorporate national, international, and industry-consensus best practices. Whenever a company updates those safety protocols, it has 30 days to publish the changes along with its rationale.

SB 53 also provides legal protections for whistleblowers exposing significant safety risks and creates a pathway for companies and the public to report critical safety incidents to the California Office of Emergency Services. The state Attorney General can enforce civil penalties for noncompliance, but notably, proposals for mandatory third-party audits did not make the final cut.

Big Tech was fiercely divided

AI developers spent months aggressively fighting state-level rules, arguing they would push startups and tech hubs out of California.

Anthropic eventually endorsed SB 53 after lawmakers negotiated changes to the bill's wording. Meta took a more aggressive stance, launching a state-level super PAC in August to counter California's AI push.

OpenAI also lobbied hard against it. Chris Lehane, OpenAI's chief global affairs officer, urged Newsom to let frontier labs bypass California-specific rules if they already comply with broader frameworks like the EU Code of Practice or federal agreements.

Why it matters

For researchers and tech workers, the real victory here is whistleblower protection. Exposing catastrophic flaws in frontier models has historically been a career-ending move; SB 53 gives employees legal cover and an official channel with the state's emergency office.

For the broader industry, California just proved that AI legislation can survive Big Tech lobbying if it pivots from strict liability to enforced transparency. The approach is clever because it avoids picking winners or drawing arbitrary compute thresholds that quickly become obsolete. However, without mandatory third-party audits, requiring labs to "disclose their own best practices" risks turning compliance into a PR writing contest, leaving the door wide open for future federal clashes.