Gavin Newsom, the Governor of California, issued a new executive order on Friday to expedite the implementation of two recently signed AI laws. This order addresses proposals that include mandating the development of a "kill switch" for advanced AI models.
Reasons for Issuing the Executive Order
This action comes after Newsom stated in a recent interview that he is exploring various options for further action on AI safety. He specifically referenced security concerns raised by AI researchers and emphasized the need to create legal frameworks to address the risks posed by this technology. Newsom also criticized the lack of federal action in this area, stating that former President Donald Trump dismissed warnings about the dangers of AI as a "hoax."
Details of the Executive Order
Newsom's executive order accelerates the implementation of two recently signed California laws. These laws, SB 813 and AB 1405, address how AI programs are evaluated by external groups for safety. Additionally, this order invites experts to provide recommendations to California within two months to strengthen existing AI safety and security laws. One of the proposals that may be considered in this review is updating safety law definitions to include incidents where models lose control.
Newsom has also emphasized the necessity of creating a kill switch for advanced AI models. This idea has recently been raised in Congress but has been blocked in the Senate. Some technology industry advocates, such as Dario Amodei, CEO of Anthropic, oppose this idea, arguing that kill switches are not technically feasible.
Newsom had also vetoed a bill in 2024 that proposed pre-release safety testing and a kill switch for large AI models. Instead, he tasked a panel of experts to recommend a solution. The result of this action is the SB 53 law, authored by Senator Scott Wiener, which was passed last year. This law has positioned California as a national leader in AI regulations.
However, AI safety advocates and some state lawmakers have warned that SB 53 cannot mandate the disclosure of automated hacks like the Hacking Face incident. This highlights the need for further and more effective actions regarding AI safety and oversight of this technology.
