Archive · 20 Sep 2026
California Moves Toward AI Kill Switch and Safety Oversight
California’s governor wants to make powerful artificial intelligence systems safer.
He signed an order asking experts to suggest stronger safety rules.
One idea is a kill switch that could turn off a company’s own AI system during an emergency.
Another idea is having independent groups check whether AI companies are following safety plans.
The order does not require companies to build kill switches right now.
A working group is supposed to explain how such a rule might work by Nov.
16.
Turning off AI can be difficult because models may run on many computers or be copied.
Open-weight models may also be run by people the original company cannot control.
California is also moving its AI auditor oversight program forward to the end of 2027.
Gov. Gavin Newsom signed an executive order directing experts to recommend stronger California AI safety rules.
The working group is expected to recommend how a possible frontier-model kill-switch requirement could work by Nov. 16.
The order also asks officials to consider independent third-party reviews of AI companies’ safety plans.
California’s AI auditor oversight program would begin by the end of 2027 instead of 2029.
The order studies potential requirements but does not immediately mandate kill switches for AI models.
- Who
- California Gov. Gavin Newsom, state officials, AI experts, and developers of advanced AI models.
- What
- Newsom issued an executive order directing experts to recommend stronger AI safety rules, including possible kill-switch and independent-review requirements.
- When
- Friday, Feb. 13, 2026; recommendations are requested by Nov. 16, and auditor oversight is scheduled to begin by the end of 2027.
- Where
- California.
- Why
- The order responds to concerns about AI safety and seeks faster, responsible oversight of powerful AI systems.
Supporters of stronger oversight
Implementation concerns
Whether to require kill switches
Supporters of stronger oversight
Supporters argue that developers should be able to deactivate their own powerful AI models during emergencies.
Implementation concerns
Critics of a broad interpretation caution that a developer-controlled switch cannot reliably stop copies, fine-tuned versions, or open-weight models running elsewhere.
Speed of regulation
Supporters of stronger oversight
Newsom’s administration says California should move thoughtfully but urgently because the stakes are too high to delay action.
Implementation concerns
The order itself does not impose the proposed rules immediately; it commissions recommendations so their practical design can be examined first.
California’s regulatory role
Supporters of stronger oversight
Because many leading AI companies are based in California, supporters say state rules could establish a wider industry standard.
Implementation concerns
Industry arguments cited in the article have previously characterized broad AI safety legislation as premature and too broad.
- Official
- Gavin Newsom, governor of California
- Action
- Signed an executive order on artificial intelligence safety
- Recommendation deadline
- The working group is expected to recommend how potential measures could work by Nov. 16
- Possible requirement
- Developers of frontier models could be required to build a system for deactivation during emergencies
- Independent oversight
- The order asks officials to study independent third-party reviews of companies’ safety plans
- Auditor program
- California’s AI auditor oversight program is being moved from 2029 to the end of 2027
- Current legal status
- The order studies potential requirements but does not immediately impose a kill-switch mandate
