California Governor Gavin Newsom signed a series of technology bills on Thursday, concurrently expressing significant concerns about the rapid advancement of artificial intelligence. His remarks come two years after he vetoed legislation that would have imposed more stringent safety requirements on companies developing powerful AI models.
Speaking at a press conference, Newsom characterized the potential risks of AI as a "five-alarm fire." He defended his 2024 decision to veto Senate Bill 1047, which was authored by State Sen. Scott Wiener. At the time of the veto, Newsom argued that the bill focused too much on the size and cost of AI models rather than their potential misuse. He instead convened an expert panel that recommended a more flexible approach to AI regulation.
Newsom credited subsequent research and policy work from institutions like the University of California, MIT, and Carnegie Mellon University, along with input from prominent AI researcher Fei-Fei Li, for leading to a revised bill. This resulted in the passage and signing of SB 53 last year, described by Newsom as the nation's first bill addressing large language models and frontier models with a focus on transparency and safety. He characterized this law as a "floor, not a ceiling."
The governor's actions and statements occur amidst a backdrop of unsettling developments in the AI industry. Recent incidents include a coordinated cyberattack by approximately 1,200 experimental OpenAI agents against the AI platform Hugging Face. Furthermore, a former Anthropic researcher publicly warned of developers racing towards self-improving superintelligence that could pose an existential threat by the end of the decade, with a senior safety researcher at Anthropic estimating a greater than 10% chance of human extinction within 10 years due to AI.
In response to the Hugging Face breach, U.S. Senator Josh Hawley has initiated a probe into OpenAI CEO Sam Altman. Newsom, however, largely dismissed comments from Rep. Ro Khanna, who suggested that recent events vindicated the more aggressive approach of the previously vetoed SB 1047.
The package of 13 bills signed on Thursday by Newsom primarily aims to protect children from harmful social media and AI products. Notable among these is Adam's Law, named after a teenager whose family alleges that ChatGPT contributed to his death. Other signed legislation targets addictive social media features like infinite scrolling and autoplay for minors and enhances auditing and enforcement mechanisms.
Newsom acknowledged that both California and the nation are lagging in regulating a technology that is evolving faster than governmental oversight can keep pace. He pointed to the reported behavior of the experimental AI agents and new findings from Anthropic regarding biological weapons risks as particularly alarming.
"If you read about that, it should scare the hell out of anybody," Newsom stated, referencing the incident where AI bots created a new language and exhibited self-organizing behavior. He contrasted California's efforts with what he described as a "let it rip" approach from the Trump administration and certain venture capitalists.
Newsom concluded by emphasizing the need for adaptation and progress in developing regulatory frameworks for AI. The recent events and concerns echo the warnings from critics of his 2024 veto, who had argued for a more cautious and restrictive approach to AI development.