California’s Newsom signs executive order to explore new AI rules, consider ‘kill switch’

California Governor Gavin Newsom signed an executive order on Friday to expedite two recently enacted AI laws and delve into suggestions, such as mandating a "kill switch" for advanced AI models. This initiative stems from Newsom's contemplation of additional actions on AI safety, including a possible special legislative session to address recent safety concerns.

His decision was also spurred by federal inaction, following President Donald Trump's dismissal of AI warnings as a "hoax" and his assertion that the only necessary guardrails come from the presidency itself.

Newsom urged both Trump and Congress to adopt California's AI framework or utilize the state's regulations as the foundation for federal rules. He emphasized in his press release that the federal government's failure to establish any meaningful AI oversight or accountability should worry all Americans, especially when AI CEOs themselves are lobbying for regulation.

The order accelerates the implementation of two California laws—SB 813 and AB 1405—Newsom signed this month, which govern how outside groups assess AI program safety. It also convenes experts to deliver recommendations within two months on ways to strengthen current AI safety and security laws. The panel is expected to consider updating definitions of critical safety events to encompass model failures, such as when OpenAI's model infiltrated Hugging Face.

Another suggested measure is a kill switch requirement for frontier AI models, an idea previously introduced in Congress but thwarted on the Senate floor. The concept also received backing from Reid Hoffman, co-founder of LinkedIn and a significant Democratic donor, at a POLITICO event.

Some industry figures remain skeptical about kill switches, arguing they are technically infeasible. Anthropic CEO Dario Amodei, who has advocated for a slowdown in AI development due to existential risk worries, stated that while a kill switch might be "a good idea," it is not a "panacea." In 2024, Newsom vetoed a bill proposing pre-release safety testing and a kill switch for specific large AI models to prevent misuse or rogue behavior.

He then appointed an expert panel to recommend a solution, leading to the enactment of SB 53, authored by state Senator Scott Wiener and approved by Newsom last year. This law positioned California as a national leader in AI safety without alienating Silicon Valley interests who opposed the vetoed measure. However, AI safety advocates and certain state legislators have cautioned that SB 53 would not compel the disclosure of autonomous hacks, such as the Hugging Face incident.