Newsom's kill switch order accelerates California's AI auditor registry by 13 months

California's AI auditor registry and oversight program was due to start in January 2029. The order moves it to December 2027, thirteen months earlier ^[4][5], and Newsom said it also speeds up implementation of recently enacted state laws that set a framework for independent organizations to assess AI systems for safety and risk ^[25]. Those are dates a compliance budget can be built against. The kill switch got a panel and a two-month deadline ^[2].

The publishers do not agree on what the order asks developers to build. Fox Business reported that the order requires companies developing AI to create a kill switch for their frontier models in case they go rogue ^[6]. The Indian Express described the panel as directed to consider one option among several ^[1]. Newsom's own account is the narrower one: in a video on his personal YouTube channel he said a kill switch "means a lot of things depending on who you talk to," and added, "So we want to flesh out exactly what that means and how to come up with a framework that works" ^[8][9]. He said the concept is still largely in its infancy ^[7].

The specification problem is real. The Indian Express reported that a majority of experts believe AI kill switches are impractical and would be ineffective at preventing catastrophic harm, because the compute sits across multiple data centres around the world, equipped with backup systems built to preserve workloads through an outage ^[12][13]. A switch on the software side would hand attackers a new target ^[14]. Power grids and financial systems are wired together closely enough that an abrupt shutdown could cause disruption on its own ^[15].

The audit provision is the part a company can cost out. The order urges the expert group to draft laws requiring independent third-party organizations to write comprehensive safety plans for AI companies, and to embed those organizations on site to conduct safety audits ^[11]. Auditor hours, a data room, a registry filing due by December 2027 ^[4].

Federal action ran the other way. The bipartisan Kill Switch Act, which would oblige frontier labs including OpenAI, Anthropic, Google and Meta to build a fast shutdown mechanism and would give the Department of Homeland Security authority to use it, is stalled in the House ^[16]. Sen. John Kennedy tried to introduce a federal kill switch bill this week and Sen. Rand Paul blocked it ^[17]. "I think it would be crazy to get on the floor one day and regulate an entire industry," Paul told reporters ^[18]. Newsom, in his statement, said "The federal government's abject failure to create any form of meaningful AI oversight or accountability should alarm every American, especially when AI CEOs themselves are begging for regulation" ^[10].

So the panel reports around mid-November ^[26] with a definition tight enough to legislate, and Newsom, who has floated a special session on AI ^[21], puts a shutdown duty into statute. Or it reports a framework that restates the engineering objections, and the obligations with dates stay the registry and the outside assessors. Or Congress moves and the state question narrows. I'd expect the second, because the veto nearly two years ago killed a mandate ^[3] and this order sends the same mandate to a study. A special session bill that names a shutdown duty and a date would settle it the other way.