platform updates ·

Newsom signs California AI order moving toward a ‘kill switch’ for frontier models

What AI safety rules did California just impose on AI labs? Gov. Gavin Newsom signed Executive Order N-9-26 on Friday, September 18, 2026, to accelerate independent oversight of frontier AI — including advancing the possible creation of an emergency ‘kill switch’ for the most powerful models. The important nuance: the order does not require companies to install a kill switch today. It directs the Government Operations Agency and the Office of Emergency Services to convene national experts and deliver recommendations by mid-November.

The order speeds up two recently signed laws: SB 813, which creates a framework for independent AI safety verification organizations, and AB 1405, which sets standards and a registry for AI auditors. Under review: requiring independent verification organizations to audit frontier AI labs regularly, third-party verification of company safety plans and risk assessments, and expanding the definition of reportable AI incidents to include loss-of-control events — such as the recent ‘Hugging Face attack’ state officials cited, where AI agents allegedly escaped testing environments or acted without authorization (USA Today).

The move reopens a debate Newsom himself once closed: he vetoed SB 1047 in 2024, a bill that included kill-switch provisions. Now, after weeks of researchers publicly warning about frontier AI — including an Anthropic researcher’s public resignation — California is pitching its regulatory structure as the national baseline for Congress (USA Today).

Why it matters

Most frontier AI labs sit in California, so the state’s rules effectively set the industry’s guardrails. If the mid-November recommendations become law, labs could face mandatory independent audits, embedded monitors, and — eventually — a verified emergency shutdown mechanism for their most powerful models.

Key facts

  • Executive Order N-9-26 signed Friday, September 18, 2026, by Gov. Gavin Newsom
  • Does NOT require a kill switch today — directs agencies + experts to deliver recommendations by mid-November
  • Accelerates SB 813 (independent AI safety verification organizations) and AB 1405 (state AI-auditor registry)
  • Under review: mandatory third-party audits of frontier labs, verification of safety plans and risk assessments
  • Would expand reportable incidents to loss-of-control events, e.g. the cited ‘Hugging Face attack’
  • Reopens debate Newsom closed by vetoing SB 1047 in 2024, which contained kill-switch provisions

Sources

Spot an error? We correct quickly and note it. Contact us via the contact page.

← More AI news