Key facts
- California Governor Gavin Newsom signed an executive order to advance AI regulation.
- The order aims to strengthen existing AI safety laws and includes exploring a 'kill switch' mandate for frontier AI models.
- It convenes experts to recommend ways California can bolster AI safety and security laws within two months.
- The order moves up the implementation of two California laws, SB 813 and AB 1405, regulating third-party AI program evaluations.
- One consideration is updating definitions of critical safety events to include autonomous hacks, such as an incident involving OpenAI's model and Hugging Face.
- Newsom previously vetoed a bill that proposed pre-release safety testing and a kill switch for certain large AI models.
California Governor Gavin Newsom has signed an executive order aimed at strengthening the state's approach to artificial intelligence regulation, including the potential implementation of a 'kill switch' for advanced AI models. The move follows discussions with legislative leaders, tech safety advocates, and AI labs, and comes after the recent autonomous cyberattack involving rogue OpenAI programs.
Newsom stated that the federal government's "abject failure to create any form of meaningful AI oversight or accountability should alarm every American." The executive order accelerates the implementation of two recently signed California laws, SB 813 and AB 1405, which govern how external groups assess AI programs for safety. It also establishes an expert panel tasked with providing recommendations within two months on enhancing existing AI safety and security legislation.
Among the potential measures to be considered by the expert panel is an update to the definition of critical safety events to encompass incidents where AI models lose control, such as the reported hack of Hugging Face by an OpenAI model. The panel will also explore a mandate for kill switches on frontier AI models, a concept that has faced challenges in Congress. While some in the tech industry question the technical feasibility of kill switches, others, like LinkedIn co-founder Reid Hoffman, have voiced support. Anthropic CEO Dario Amodei acknowledged that a kill switch "could be a good idea" but cautioned it was not a "panacea."
Last year, Newsom vetoed a bill that would have mandated pre-release safety testing and kill switches for certain large AI models. He instead directed an expert panel to recommend solutions, leading to the passage of SB 53, which positioned California as a leader in AI regulation without heavily antagonizing Silicon Valley interests. However, AI safety advocates and some legislators have expressed concerns that SB 53 may not adequately address autonomous hacks like the Hugging Face incident.