California state senator makes second play for guardrails on AI

SACRAMENTO, Calif. (CN) — The California state senator who had a major bill on artificial intelligence vetoed last year has introduced new legislation focused on the emerging technology.

State Senator Scott Wiener, a San Francisco Democrat, returned this week with Senate Bill 53. It’s a placeholder bill, as specific details aren’t yet listed, though they can be added later.

The senator, in an apparent follow-up to the vetoed Senate Bill 1047 from last year, in his new bill says that it’s the Legislature’s intent to create safeguards for developing AI frontier models. Those safeguards would build state capacity for using AI that could include any findings by the Joint California Policy Working Group on AI Frontier Models created by Governor Gavin Newsom.

The Legislature convened on Monday for the first time this year. Wiener’s bill hasn’t yet been assigned to a policy committee.

“My office is considering several AI proposals for this session, some that pertain to safety and some that do not,” Wiener said in a statement. “I’m planning to introduce a full proposal in the next month or two. I’m also closely watching the work of the governor’s AI working group, which may lead to more ideas and opportunities for collaboration.”

Vetoing Wiener’s previous bill in September, Newsom said it fell short of a flexible, comprehensive solution. That bill would have required AI developers, along with the people who provide the computing power that trains them, to ensure catastrophic harm was avoided by placing guardrails on AI.

Newsom in his veto message said the bill would have created a framework that could give people a false sense of security.

“While the large AI labs have made admirable commitments to monitor and mitigate these risks, the truth is that voluntary commitments from industry are not enforceable and rarely work out well for the public,” Wiener said in a statement at the time.

Newsom announced a new working group at the same time he vetoed Wiener’s bill — the Joint California Policy Working Group on AI Frontier Models. The governor said he asked several AI experts to sit on the panel and help develop rules for the technology, which will focus on analyzing the capabilities and risks of frontier AI models. Newsom indicated he’d work with lawmakers this year on the issue.

The group — which includes “godmother of AI” Fei-Fei Li, co-director of the Stanford Institute for Human-Centered Artificial Intelligence — has started writing a draft report. It will detail the latest information about frontier model capabilities and risks and is expected before April.

People with a variety of expertise will receive the report, which is meant to promote feedback. In-person and remote workshops are a possibility. Those interested can then give written comment about the draft report.

Adding that feedback to the document, the group anticipates completing its report by this summer.

Wiener’s AI bill from last year was one of the most high-profile pieces of legislation, leading Elon Musk — owner of SpaceX, Tesla and social media giant X — to publicly weigh in with support.

The senator in August, days before a crucial Assembly committee vote, hosted a series of AI experts to talk about the need for guardrails.

Experts at that event warned that AI hadn’t yet reached a point where it could assist someone in making a pandemic-level virus. However, the possibility was on their minds.

Kevin Esvelt, associate professor at MIT Media Lab and co-founder of Secure Bio, said that while the risk is small, it isn’t zero percent and the vulnerability should be closed. AI continues to advance, and he theorized it could have the ability, for example, to help someone shut down the electrical grid in two years.

Esvelt added that technology shouldn’t be created before a defense exists for it.

That was one reason why Ari Kagan, an advisor to Economic Security California Action, supported the legislation. Kagan said it would have required testing AI to determine if it had a capability someone could misuse.