California’s next major AI policy fight may be narrower than the last one, but it is still aimed squarely at the frontier of artificial intelligence. State Senator Scott Wiener has introduced SB 53, a new bill that would protect workers at leading AI labs when they raise concerns about dangerous systems and would begin the process of creating a public cloud computing cluster called CalCompute.
The proposal follows the collapse of SB 1047, Wiener’s earlier AI safety bill, which Governor Gavin Newsom vetoed in September. SB 53 does not try to recreate the full structure of that bill. Instead, it focuses on two pieces that were among the least controversial: whistleblower protections and public access to compute.
A narrower bill after SB 1047
SB 1047 became one of the central AI policy fights of 2024. The bill was designed to reduce the chance that very large AI models could contribute to catastrophic events, including loss of life or cyberattacks with more than $500 million in damages.
That approach drew intense criticism from parts of Silicon Valley. Some leaders argued that SB 1047 could weaken America’s competitive position in the global AI race. Others framed the bill as a response to unrealistic fears about AI systems and doomsday scenarios.
Wiener pushed back against that criticism. He alleged that some venture capitalists had mounted a “propaganda campaign” against SB 1047, pointing in part to Y Combinator’s claim that the bill would send startup founders to jail. Experts cited in the source argued that claim was misleading.
Governor Gavin Newsom ultimately vetoed SB 1047 in September, saying it was not the best approach. SB 53 appears to be a more targeted response to the same broad concern: how to govern powerful AI systems without relitigating every part of the earlier bill at once.
What SB 53 would protect
The core of SB 53 is aimed at employees inside major AI companies. The bill would protect workers who believe their employer’s AI systems could create a critical risk to society and who disclose concerning information to California’s Attorney General, federal authorities, or other employees.
The bill defines critical risk around severe outcomes. The definition includes foreseeable or material risks tied to the development, storage, or deployment of a foundation model that could result in death or serious injury to more than 100 people, or more than $1 billion in damage to rights in money or property.
That language shows that Wiener is not avoiding existential AI risk in the new proposal. Even though SB 53 is more limited than SB 1047, it still treats the possibility of extreme harm from foundation models as a legitimate policy issue.
The whistleblower protections would apply to frontier AI model developers. The source says that category would likely include OpenAI, Anthropic, and xAI, among others. These developers would be limited from retaliating against employees who make protected disclosures.
SB 53 would also require covered developers to report back to whistleblowers on certain internal processes that the workers identify as concerning. That provision matters because it moves the bill beyond simply allowing employees to speak. It also creates a channel for internal concerns to receive a response.
CalCompute would expand access to AI infrastructure
The second major piece of SB 53 is CalCompute, a proposed public cloud computing cluster. The stated purpose is to give researchers and startups access to computing resources needed to develop AI that benefits the public.
Rather than immediately setting every detail, SB 53 would establish a group to work out how CalCompute should be built. That group would include University of California representatives along with other public and private researchers.
The group would make recommendations on several practical questions:
- How CalCompute should be built.
- How large the cluster should be.
- Which users and organizations should be allowed to access it.
This part of the bill sits differently from the whistleblower provisions. While the worker protections are about risk, CalCompute can also be seen as a pro-innovation measure. It is meant to broaden access to AI development resources beyond the companies that already have large computing capacity.
That dual structure may be important for the politics of SB 53. The bill combines a safety-focused mechanism with an infrastructure proposal that could help researchers and startups build AI systems for public benefit.
The political test ahead
SB 53 is still early in the legislative process. It must be reviewed and passed by California’s legislative bodies before it can reach Governor Newsom’s desk. State lawmakers are also expected to watch how Silicon Valley responds.
The environment in 2025 may be more difficult for AI safety legislation than it was in 2024. California passed 18 AI-related bills in 2024, but the source notes that the AI doom movement appears to have lost ground.
National signals may also shape the debate. Vice President J.D. Vance signaled at the Paris AI Action Summit that America is more focused on AI innovation than AI safety. That could make the safety side of SB 53 harder to sell, even if CalCompute can be framed as supporting AI progress.
The bill’s prospects may depend on whether lawmakers and industry leaders see it as a modest correction after SB 1047 or as a continuation of the same fight under a new name. Its narrower design gives supporters a simpler case to make: protect employees who raise serious warnings and build public AI infrastructure.
Even so, SB 53 keeps the central question alive. California is still wrestling with how to address the risks of powerful AI systems while maintaining room for startups, researchers, and major labs to keep building. The answer will now move through another round of legislative scrutiny.