California senator introduces new AI safety bill SB 53

Silicon Valley's most controversial AI safety debate isn't dead after all.

California state Senator Scott Wiener, the architect behind last year's highly contested SB 1047, is back with a fresh legislative push. On Friday, Wiener introduced SB 53, a new bill that essentially strips away the most toxic provisions of its predecessor and repackages the parts that actually had broad support: legal protections for AI whistleblowers and public compute access for researchers.

Legal cover for AI whistleblowers

Under SB 53, frontier AI developers like OpenAI, Anthropic, and xAI would be strictly prohibited from retaliating against employees who raise red flags about models that pose a critical risk to society. Staff members could report their concerns directly to California's Attorney General, federal regulators, or even discuss them internally without fear of termination or lawsuits. The bill would also force companies to report back to whistleblowers on specific internal processes under scrutiny.

Wiener isn't backing down from catastrophic risks either, though he is redefining the line. The new text sets critical risk as a foreseeable danger that could result in more than 100 deaths or severe injuries, or exceed $1 billion in monetary damages. That effectively doubles the financial damage threshold from SB 1047's original $500 million benchmark.

A public cloud for the rest of us

To balance the stick with a carrot, SB 53 also revives CalCompute, an initiative to build a state-backed public cloud computing cluster.

A new advisory group consisting of University of California representatives along with public and private researchers would determine the cluster's size, architecture, and who actually gets access. The goal is to give university labs and underfunded startups the raw computing muscle required to train models without relying entirely on tech giants.

Why it matters

Wiener is playing a much smarter political game this time around. SB 1047 exploded in 2024 because VCs and tech leaders painted it as an innovation-killing machine, with Y Combinator famously warning that founders could face criminal liability before Governor Gavin Newsom ultimately vetoed it. By zeroing in on whistleblower protections and shared infrastructure, Wiener makes SB 53 significantly harder for tech lobbyists to attack without looking like they want to silence their own scientists.

For engineers and researchers inside frontier labs, this bill could be a massive deal. Over the past year, we have seen high-profile safety researchers walk away from top labs over restrictive non-disparagement clauses and ethical disputes. Legal immunity would completely change their leverage.

Still, passing any AI safety bill in 2025 faces steep headwinds. The national narrative has shifted aggressively toward maintaining America's lead over competitors. Vice President J.D. Vance recently signaled at the Paris AI Action Summit that the administration prioritizes AI innovation over safety restrictions. While California passed 18 AI-related measures in 2024, the appetite for catastrophic-risk legislation is clearly waning, making SB 53's path through the legislature far from guaranteed.