California Governor Gavin Newsom has officially vetoed Senate Bill SB-1047, which aimed at regulating large AI models. The bill contained numerous unreasonable provisions that could have severely impacted both large companies and startups. Governor Newsom highlighted that a blanket approach like SB-1047 could stifle innovation and the development of AI technology in certain fields, and its regulations might be difficult to implement.
Over the past 30 days, Governor Newsom has signed 17 bills related to regulating large AI models and ensuring the safe use of generative AI, but SB-1047 is the only one he vetoed. Key figures in the AI community, such as Andrew Ng, Yann LeCun, and Fei-Fei Li, played significant roles in opposing the bill, especially through public calls for tech professionals to resist it.
The veto of SB-1047 is a significant win for global developers, allowing continued use of open-source models from major U.S. tech companies like Meta and Google.
SB-1047, officially drafted on February 7, aimed to enhance the safety, transparency, and usage standards of large AI models. However, it contained many unreasonable clauses, such as penalizing developers if their models, costing over $100 million to develop, were misused. Additionally, the bill required companies to submit detailed customer information and conduct annual user behavior evaluations, with data backup for seven years and customs oversight, which many saw as overreach.
The bill's implementation could have devastated large companies like OpenAI, Meta, and Google, as well as smaller startups. California, being a major tech hub housing the headquarters of companies like Google, Meta, OpenAI, Apple, Intel, and Tesla, and top universities like Stanford and UC Berkeley, would have faced potential exodus of tech firms.
Governor Newsom stated that California hosts 32 of the world’s top 50 AI companies crucial for AI model development and innovation. While SB-1047 was well-intentioned, it focused solely on expensive, large-scale AI models, possibly giving a false sense of security in controlling rapidly evolving technology. Specialized models could be equally or more dangerous. The bill also lacked flexibility for different AI applications, which could cause confusion and uncertainty in its implementation.
Newsom emphasized that SB-1047 did not consider whether AI models were deployed in high-risk environments, nor did it clearly define critical decisions or sensitive data, potentially compromising privacy and data security. Effective protection from AI threats requires more nuanced, targeted regulation rather than a one-size-fits-all approach.
