California has officially enacted Senate Bill 53, a significant piece of legislation aimed at enhancing the transparency of artificial intelligence (AI) systems. The law is the culmination of extensive debates and discussions within the tech community.
On Monday, Governor Gavin Newsom signed the “Transparency in Frontier Artificial Intelligence Act,” drafted by Senator Scott Wiener (D-CA). This marks the second iteration of the bill, following the veto of the original version, SB 1047, last year. Concerns were raised at that time about the bill being overly restrictive and potentially hindering AI innovation within the state. The original proposal would have mandated that AI developers, particularly those with training costs exceeding $100 million, conduct assessments for specific risks. After the veto, Newsom directed AI researchers to propose an alternative approach, resulting in a comprehensive 52-page report that informed the creation of SB 53.
Key recommendations from the researchers were incorporated into SB 53. Among these is the requirement for large AI firms to disclose their safety and security protocols. The law also establishes protections for whistleblowers within these companies and mandates direct public information sharing to enhance transparency. However, some recommendations, such as the implementation of third-party evaluations, were not included.
Under the terms of the new legislation, major AI developers must make publicly available a framework detailing how they integrate national and international standards, as well as industry consensus best practices into their AI systems. Any updates to their safety protocols must be published, along with the rationale for those changes, within a 30-day timeframe. This aspect of the legislation may not fully satisfy advocates for stronger AI regulation, as many companies favor voluntary guidelines that lack substantial penalties.
Moreover, the legislation introduces a mechanism for both AI companies and the public to report serious safety incidents to the California Office of Emergency Services. It also includes protections for whistleblowers disclosing significant health and safety risks associated with AI models, alongside civil penalties for noncompliance, enforceable by the Attorney General’s office. The California Department of Technology will be tasked with recommending updates to the law annually, incorporating feedback from multiple stakeholders and responding to advancements in technology and international standards.
The response from AI companies has been mixed. Many initially opposed SB 53, fearing it could push businesses out of California. The stakes are high, as California’s large population and emerging AI hubs significantly influence the industry and its regulatory landscape.
While SB 53 secured public support from Anthropic following weeks of negotiations, Meta took steps to influence AI policy by launching a state-level super PAC in August. OpenAI’s lobbying efforts were directed against such regulations, with its global affairs officer, Chris Lehane, conveying to Newsom that effective technology regulation is best achieved when it complements broader federal and global safety initiatives.
Lehane proposed that companies could fulfill California’s requirements by adhering to federal or global agreements, advocating for compliance recognition when developers engage with established regulatory frameworks such as the EU Code of Practice or enter into safety agreements with relevant U.S. federal agencies.
