California’s state senate has given final approval to a significant AI safety bill, SB 53, which aims to establish new transparency requirements for large artificial intelligence companies. Authored by state senator Scott Wiener, the bill encompasses several key provisions designed to enhance accountability and safety within the rapidly evolving AI sector. Specifically, SB 53 mandates that large AI laboratories disclose their safety protocols, provides whistleblower protections for their employees, and establishes a public cloud initiative known as CalCompute to broaden access to computing resources.

The bill now proceeds to California Governor Gavin Newsom for his signature or veto. Newsom's stance on AI legislation has been cautious in the past; last year, he vetoed a more expansive safety bill, also from Senator Wiener, while simultaneously approving narrower legislation that addressed issues like deepfakes. At the time of his previous veto, Newsom acknowledged the critical need to “protecting the public from real threats posed by this technology.” However, he expressed concerns that Wiener's earlier bill applied “stringent standards” indiscriminately to large models, regardless of whether they were deployed in high-risk environments, involved critical decision-making, or utilized sensitive data. Senator Wiener indicated that the current iteration of SB 53 was developed with recommendations from a policy panel of AI experts convened by Governor Newsom himself after the previous veto, suggesting an attempt to address those earlier criticisms.

Recent amendments to SB 53 also reflect a nuanced approach to regulation. Companies developing “frontier” AI models, defined by generating less than $500 million in annual revenue, will be required to disclose high-level safety details. In contrast, companies exceeding this revenue threshold will be mandated to provide more comprehensive and detailed reports, creating a tiered system of compliance.

Despite its passage through the senate, SB 53 has faced considerable opposition from various entities within Silicon Valley, including numerous venture capital firms and lobbying groups. OpenAI, in a letter to Governor Newsom, argued against potential “duplication and inconsistencies” and suggested that companies meeting federal or European AI safety standards should be deemed compliant with statewide rules. Andreessen Horowitz’s head of AI policy and chief legal officer raised constitutional concerns, asserting that many state AI bills, including proposals in California and New York, risk violating constitutional limits on state regulation of interstate commerce. This sentiment aligns with previous statements from a16z’s co-founders, who have cited tech regulation as a reason for supporting Donald Trump’s political aspirations, an administration that subsequently called for a decade-long ban on state-level AI regulation.

Conversely, the AI company Anthropic has publicly supported SB 53. Jack Clark, co-founder of Anthropic, stated that while the company would prefer a federal standard for AI governance, in its absence, SB 53 represents a “solid blueprint” that merits serious consideration. This highlights the divided opinions within the tech industry regarding state-level AI regulation and the path forward for ensuring responsible AI development and deployment.