SACRAMENTO, CA – The California State Legislature has passed a landmark bill that could establish the most stringent artificial intelligence safety regulations in the United States. The bill, SB 1047, mandates that developers of the most powerful AI systems, referred to as “frontier models,” adhere to strict safety testing protocols, implement robust security measures, and assume liability for potential harms.
The legislation, authored by State Senator Scott Wiener, narrowly passed the State Assembly and now heads to Governor Gavin Newsom’s desk for his signature or veto. If signed into law, it would require AI companies to perform comprehensive risk assessments and certify to the state that their models cannot be used for catastrophic purposes, such as creating biological or chemical weapons. The bill also contains a provision requiring developers to create a “kill switch” to shut down a model if it exhibits uncontrollable or dangerous behaviors.
A key component of SB 1047 is the establishment of a new state entity, the Frontier Model Division, which would operate under the Department of Technology. This body would have the authority to oversee the development of large-scale models trained with significant computational power (defined as those using 10^26 integer or floating-point operations).
The bill has ignited a fierce debate between AI safety advocates and the tech industry. Supporters, including the Center for AI Safety, argue that the legislation provides necessary guardrails to prevent potential AI-driven disasters. However, tech industry groups, including TechNet—which represents giants like Google and Meta—have strongly opposed the bill. They contend that its rigid requirements and high compliance costs will stifle innovation, drive startups out of California, and put the state at a competitive disadvantage. The outcome of Governor Newsom’s decision is now being closely watched as it could set a powerful precedent for AI governance across the nation.
English 
Deutsch