California lawmakers pass AI safety bill SB 53 — but Newsom could still veto

In a significant move early Saturday, California's state senate approved a landmark AI safety bill aimed at increasing transparency among major technology firms. Authored by state senator Scott Wiener, the legislation, known as SB 53, mandates that large AI laboratories disclose their safety protocols, introduces protections for whistleblowers within these firms, and establishes a public cloud resource, dubbed CalCompute, to enhance computing access. The bill is now presented to Governor Gavin Newsom, who has yet to provide public commentary on the legislation. Notably, Newsom vetoed a more comprehensive safety measure proposed by Wiener last year, while opting to sign a more focused bill addressing issues such as deepfakes. During that time, he recognized the necessity of safeguarding the public from the genuine risks associated with AI technology, yet he criticized the previous proposal for imposing excessive standards on large models without considering their operational context or the sensitivity of the data involved. Wiener has indicated that the current bill reflects insights from an advisory panel of AI experts that Newsom had assembled following the prior veto. Recent amendments to SB 53 stipulate that companies developing advanced AI models with annual revenues under $500 million will only be required to provide general safety information, whereas those exceeding that revenue threshold will need to submit more detailed disclosures. The bill has faced opposition from several prominent Silicon Valley companies, venture capitalists, and lobbying organizations. In a recent correspondence to Newsom, OpenAI refrained from directly addressing SB 53 but suggested that to prevent overlapping regulations, firms should be deemed compliant with state safety standards provided they adhere to federal or European guidelines. Additionally, a chief legal officer from Andreessen Horowitz expressed concerns that numerous state-level AI regulations, including those in California and New York, may overstep constitutional boundaries regarding interstate commerce. On the other hand, Anthropic has voiced its support for SB 53, with co-founder Jack Clark stating, "While we would prefer a federal standard, this legislation offers a robust framework for AI governance that cannot be overlooked." The fate of SB 53 now hangs in the balance as it awaits the governor's decision.