California passes SB 53 AI safety bill, but a veto still looms

Silicon Valley's favorite regulatory headache is back on the governor's desk.

Early Saturday morning, the California state senate officially passed SB 53, a major AI safety bill crafted by state senator Scott Wiener. The legislation now heads to Governor Gavin Newsom, who vetoed Wiener's more sweeping AI safety bill last year but hasn't publicly commented on this latest effort.

A lighter touch aimed at the giants

According to Wiener, SB 53 zeroes in on three things: requiring large AI labs to be transparent about their safety protocols, establishing whistleblower protections for AI workers, and launching CalCompute, a public cloud meant to expand compute access for researchers.

Wiener drafted this version using recommendations from an expert panel Newsom convened after his previous veto. Last year, Newsom pushed back on Wiener's earlier bill for placing strict, blanket rules on models regardless of whether they were deployed in high-risk environments or dealt with critical decisions.

Lawmakers also added a notable compromise. As reported by Politico, an amendment sets an annual revenue threshold of $500 million: frontier model developers making under that figure only need to disclose high-level safety summaries, while tech giants crossing that line have to submit full, detailed documentation.

Big Tech draws the battle lines

The vote has split Silicon Valley into opposing camps.

Anthropic quickly came out in favor. Co-founder Jack Clark noted on social media that while the lab would prefer federal action, in the absence of a national framework, SB 53 sets a solid blueprint for AI governance that cannot be ignored.

Meanwhile, VC firms and major rivals are pushing back hard. In a letter to Newsom, OpenAI argued that to avoid inconsistent and duplicate rules, companies should automatically be deemed compliant with state laws if they already meet federal or European safety standards.

Andreessen Horowitz went even further. Its chief legal officer argued that state-level bills like SB 53 overstep constitutional boundaries by interfering with interstate commerce. Notably, a16z founders previously backed Donald Trump's reelection effort partly over tech regulation, and the Trump administration subsequently pushed for a 10-year freeze on state AI rules.

Why it matters

Wiener played this round much smarter than the last. By dropping the $500 million carve-out into the text, the bill shields early-stage startups and gives Newsom a politically viable off-ramp to claim he protected California innovation while holding dominant labs accountable.

Still, the pushback from OpenAI and a16z highlights the fundamental flaw of state-by-state AI policy. Code does not stop at state lines. Forcing frontier labs to juggle California disclosures, potential New York requirements, and EU rules will trigger massive compliance overhead and likely accelerate legal challenges over constitutional limits on interstate commerce.

For everyday developers, SB 53 offers whistleblower protections and potential access to subsidized compute through CalCompute without immediate reporting headaches. But the fate of the bill sits entirely with Newsom. If he signs it, California cements itself as America's de facto AI regulator; if he vetoes it again, he will reinforce expectations that frontier AI oversight belongs strictly at the federal level.