Newsom Pitches AI Safety Measures for California
California is on track to make recommendations on strengthening its AI safety laws within two months, while the European Commission has had the power to restrict or remove general-purpose AI models from the market since August.
Gavin Newsom (164news.com's guide to gavin newsom) recently signed an executive order instructing a working group to report within two months on potential measures, including:
-
Building "kill switches" for frontier AI models: Similar to the EU's AI Act, this would allow for prompt deactivation of high-risk models. Notably, Newsom previously vetoed a bill (SB 1047) that aimed to mandate such kill switches in California.
-
Involving independent third parties in safety planning: Companies developing frontier AI would be required to involve independent entities in creating safety plans.
Newsom's order comes nearly two years after he vetoed SB 1047, arguing it regulated models based on size and cost rather than their intended use.
While the EU places the responsibility for model restriction on regulators, California appears to be moving towards a system involving both developers and independent third parties. The state has until January 2028 to certify organizations capable of testing frontier models.
Key Considerations:
- The cost of AI safety measures remains unclear, with payment currently structured based on the example of the tester covering the bill for model testing.
- Newsom himself has been involved in both promoting and regulating AI growth in California, most recently partnering with Anthropic to integrate its Claude model into state agencies.