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

California’s state senate gave last approval early on Saturday morning to a serious AI security invoice setting new transparency necessities on giant firms.

As described by its creator, state senator Scott Wiener, SB 53 “requires giant AI labs to be clear about their security protocols, creates whistleblower protections for [employees] at AI labs & creates a public cloud to develop compute entry (CalCompute).”

The invoice now goes to California Governor Gavin Newsom to signal or veto. He has not commented publicly on SB 53, however final yr, he vetoed a extra expansive security invoice additionally authored by Wiener, whereas signing narrower laws focusing on points like deepfakes.

On the time, Newsom acknowledged the significance of “defending the general public from actual threats posed by this expertise,” however criticized Wiener’s earlier invoice for making use of “stringent requirements” to giant fashions no matter whether or not they had been “deployed in high-risk environments, [involved] crucial decision-making or using delicate knowledge.”

Wiener stated the brand new invoice was influenced by suggestions from a coverage panel of AI consultants that Newsom convened after his veto.

Politico additionally stories that SB 53 was lately amended in order that firms growing “frontier” AI fashions whereas bringing in lower than $500 million in annual income will solely must disclose excessive stage security particulars, whereas firms making greater than that might want to present extra detailed stories.

The invoice has been opposed by quite a few Silicon Valley firms, VC companies, and lobbying teams. In a current letter to Newsom, OpenAI didn’t point out SB 53 particularly however argued that to keep away from “duplication and inconsistencies,” firms needs to be thought of compliant with statewide security guidelines so long as they meet federal or European requirements.

And Andreessen Horowitz’s head of AI coverage and chief authorized officer lately claimed that ”lots of as we speak’s state AI payments — like proposals in California and New York — threat” crossing a line by violating constitutional limits on how states can regulate interstate commerce.

a16z’s co-founders had beforehand pointed to tech regulation as one of many components main them to again Donald Trump’s bid for a second time period. The Trump administration and its allies subsequently referred to as for a 10-year ban on state AI regulation.

Anthropic, in the meantime, has come out in favor of SB 53.

“We’ve lengthy stated we would like a federal normal,” stated Anthropic co-founder Jack Clark in a publish. “However within the absence of that this creates a strong blueprint for AI governance that can’t be ignored.”