California sets New Standard for AI Safety with Landmark Legislation
California governor Gavin Newsom recently signed Senate bill 53 (SB 53) into law, establishing a first-in-the-nation framework for regulating artificial intelligence. The legislation aims to increase openness and accountability within the AI sector, requiring companies developing powerful AI models to disclose information and report incidents to the state.
The bill has garnered support from major tech companies. Meta spokesperson Christopher Sgro described SB 53 as “a positive step” toward “balanced AI regulation,” adding that it would facilitate cooperation between state and federal governments on AI deployment. Similarly,a representative from Google stated the company believes the law represents an “effective approach to AI safety.”
SB 53’s impact is expected to be global, given that 32 of the world’s top 50 AI companies are based in california. The law mandates that AI firms report incidents to California’s Office of Emergency Services and provides protections for whistleblowers, enabling employees to voice safety concerns without fear of retribution. Noncompliance will be subject to civil penalties enforced by the state attorney general, though experts note thes penalties are relatively mild compared to those outlined in the EU’s AI Act.
While acknowledging the bill as “a step forward,” Miles Brundage, former head of policy research at OpenAI, emphasized the need for “actual transparency” in reporting, stronger minimum risk thresholds, and robust third-party evaluations. Collin McCune, head of government affairs at Andreessen Horowitz, expressed concern that the law “risks squeezing out startups, slowing innovation, and entrenching the biggest players,” possibly creating a complex regulatory landscape for smaller companies. Several AI companies echoed these concerns during lobbying efforts.
Though, Thomas Woodside, a cofounder at Secure AI Project – a cosponsor of the law – dismissed concerns about the impact on startups as “overblown.” He clarified that the reporting requirements primarily apply to companies training AI models requiring substantial computational resources – investments beyond the reach of most startups. “This bill is only applying to companies that are training AI models with a huge amount of compute that costs hundreds of millions of dollars, something that a tiny startup can’t do,” Woodside told Fortune. He further noted that many obligations only apply to companies with annual revenue exceeding $500 million.
