4 weeks ago
4
SACRAMENTO, Calif. -- California Gov. Gavin Newsom connected Monday signed a instrumentality that aims to forestall radical from utilizing almighty artificial quality models for perchance catastrophic activities similar gathering a bioweapon oregon shutting down a slope system.
The determination comes arsenic Newsom touted California arsenic a person successful AI regularisation and criticized the inaction astatine the national level successful a caller speech with erstwhile President Bill Clinton. The caller instrumentality volition found immoderate of the first-in-the-nation regulations connected large-scale AI models without hurting the state's homegrown industry, Newsom said. Many of the world's apical AI companies are located successful California and volition person to travel the requirements.
"California has proven that we tin found regulations to support our communities portion besides ensuring that the increasing AI manufacture continues to thrive. This authorities strikes that balance," Newsom said successful a statement.
The authorities requires AI companies to instrumentality and disclose publically information protocols to forestall their astir precocious models from being utilized to origin large harm. The rules are designed to screen AI systems if they conscionable a "frontier" threshold that signals they tally connected a immense magnitude of computing power.
Such thresholds are based connected however galore calculations the computers are performing. Those who crafted the regulations person acknowledged the numerical thresholds are an imperfect starting constituent to separate today's highest-performing generative AI systems from the adjacent procreation that could beryllium adjacent much powerful. The existing systems are mostly made by California-based companies similar Anthropic, Google, Meta Platforms and OpenAI.
The authorities defines a catastrophic hazard arsenic thing that would origin astatine slightest $1 cardinal successful harm oregon much than 50 injuries oregon deaths. It's designed to defender against AI being utilized for activities that could origin wide disruption, specified arsenic hacking into a powerfulness grid.
Companies besides person to study to the authorities immoderate captious information incidents wrong 15 days. The instrumentality creates whistleblower protections for AI workers and establishes a nationalist unreality for researchers. It includes a good of $1 cardinal per violation.
It drew absorption from immoderate tech companies, which argued that AI authorities should beryllium done astatine the national level. But Anthropic said the regulations are "practical safeguards" that marque authoritative the information practices galore companies are already doing voluntarily.
"While national standards stay indispensable to debar a patchwork of authorities regulations, California has created a beardown model that balances nationalist information with continued innovation," Jack Clark, co-founder and caput of argumentation astatine Anthropic, said successful a statement.
The signing comes aft Newsom past twelvemonth vetoed a broader mentation of the legislation, siding with tech companies that said the requirements were excessively rigid and would person hampered innovation. Newsom alternatively asked a radical of respective manufacture experts, including AI pioneer Fei-Fei Li, to make recommendations connected guardrails astir almighty AI models.
The caller instrumentality incorporates recommendations and feedback from Newsom's radical of AI experts and the industry, supporters said. The authorities besides doesn't enactment the aforesaid level of reporting requirements connected startups to debar hurting innovation, said authorities Sen. Scott Wiener of San Francisco, the bill's author.
"With this law, California is stepping up, erstwhile again, arsenic a planetary person connected some exertion innovation and safety," Wiener said successful a statement.
Newsom's determination comes arsenic President Donald Trump successful July announced a program to destruct what his medication sees arsenic "onerous" regulations to velocity up AI innovation and cement the U.S.' presumption arsenic the planetary AI leader. Republicans successful Congress earlier this twelvemonth unsuccessfully tried to prohibition states and localities from regulating AI for a decade.
Without stronger national regulations, states crossed the state person spent the past fewer years trying to rein successful the technology, tackling everything from deepfakes successful elections to AI "therapy." In California, the Legislature this twelvemonth passed a fig of bills to code information concerns astir AI chatbots for children and the usage of AI successful the workplace.
California has besides been an aboriginal adopter of AI technologies. The authorities has deployed generative AI tools to spot wildfires and code road congestion and roadworthy safety, among different things.
___
Associated Press newsman Matt O'Brien contributed to the report.
Copyright © 2025 by The Associated Press. All Rights Reserved.
