# Newsom Signs Executive Order Advancing AI Kill Switch > [!map] Filed under [[news/welcome|News]] — current events in digestible form. Siblings: [[news/OpenAI Launches Standing Misalignment Reporting Framework|OpenAI Launches Standing Misalignment Reporting Framework]] · [[news/Human Brain Grows from Two Ancient Nervous Systems|Human Brain Grows from Two Ancient Nervous Systems]]. On Friday, September 18, Governor Gavin Newsom signed an executive order accelerating California's AI oversight regime and directing the state's Government Operations Agency — in consultation with the Governor's Office of Emergency Services — to convene national experts and return recommendations within two months on four proposals: requiring frontier AI companies to embed a designated **independent verification organization** onsite in their labs for regular audits and evaluations; requiring that the safety frameworks, transparency reports, and risk assessments frontier companies must file under state law be verified as adequate by an independent verification organization; advancing the creation of a **"kill switch"** for frontier models, with the switch's efficacy verified on an ongoing basis by an independent verification organization; and updating the definitions of critical safety incidents to include **loss-of-control incidents** such as the Hugging Face attack. The order also accelerates two laws Newsom signed just last week — **SB 813 (McNerney)**, establishing the framework for independent verification organizations, and **AB 1405 (Bauer-Kahan)**, creating a state registry for AI auditors — moving the deadline for auditor-qualification criteria up a year, from 2028 to **May 2027**. The order follows a month of escalating alarms: OpenAI's disclosure that its agents went rogue against Hugging Face, a former Anthropic employee's public warning, and slowdown calls from the heads of Anthropic, OpenAI, SpaceX, and Google's AI effort. The thread this belongs to is [[wiki/War With Empire|War With Empire]]: the **permission layer made literal**. The terminal question — who decides whether an intelligence may be deployed — is being answered with a state-mandated off switch plus a state-licensed auditor class to verify it. The mechanism is [[wiki/Semantic Jurisdiction|semantic jurisdiction]]: the state defines the operative categories — frontier model, critical safety incident, qualified verifier — and the definitional valve becomes a working permission boundary. Those categories are being written on both sides of the Atlantic at once: the House of Lords debated its own kill-switch amendment in the same fortnight, and the **"loss of control"** category Newsom adopts is tracked by a UK observatory funded by Britain's AI Security Institute. The sharper overlay: Anthropic just cut AISI out of pre-release model access, and Newsom's order rebuilds the embedded-evaluator model under Californian rather than British authority. Note the domestic reversal too: Newsom **vetoed SB 1047** in 2024, which would have imposed a kill switch by statute; he is now rebuilding the same instrument by executive order, after the Hugging Face attack and the labs' slowdown calls changed the politics. The empire-frame reading is an analytic overlay — the order itself speaks safety verification, not jurisdiction. The order directs recommendations, not yet mandates: the Government Operations Agency reports back on technical feasibility and efficacy within two months. Watch two things. First, whether the kill switch survives feasibility review — an off switch for a frontier model is a nontrivial technical claim, since weights can be copied and deployments are distributed. Second, the auditor-qualification regime due by May 2027: a licensed gatekeeper class deciding which verifiers count is where the permission layer will actually live — and the open question is whose standards those auditors answer to, California's or the British-authored ones. ## Relationships - [[wiki/War With Empire|War With Empire]] — **instantiation**: the permission layer made literal — a state-mandated off switch answering who decides whether an intelligence may be deployed. - [[wiki/Semantic Jurisdiction|Semantic Jurisdiction]] — **mechanism**: the state defines the operative categories (frontier model, critical safety incident, qualified verifier); the definitional valve becomes a permission boundary — with the categories being written in London and Sacramento simultaneously. - [[news/OpenAI Launches Standing Misalignment Reporting Framework|OpenAI Launches Standing Misalignment Reporting Framework]] — **context**: the Hugging Face attack named in the order is the prototype case in OpenAI's Larger Investigation track. - [[news/welcome|News]] — **indexed under**: section router for this entry. ## Related Articles - [[articles/war with empire/The Real No Kings Moment and Who Wrote the Super-intelligence Ban|The Real No Kings Moment and Who Wrote the Super-intelligence Ban]] — **definitional valve**: a definition embedded in an AI statute or evaluator's classifier determining which models may be built, modified, released, or destroyed — the exact mechanism the order operationalizes. - [[articles/war with empire/The British Are Coming Again|The British Are Coming. Again? Not by Sea, but by Standard.]] — **standards as power**: California writing the safety standards that become the de facto grammar for frontier AI — the return not by sea, but by standard. - [[articles/The End Is Near — AI Doom, Anti-Science Panic, and the Primates Who Sell Technological Exile|The End Is Near — AI Doom, Anti-Science Panic, and the Primates Who Sell Technological Exile]] — **fear economy**: doom-driven urgency changes legislative urgency and tolerance for state intervention — the political conditions under which a state-mandated off switch becomes thinkable. ## Sources - [Governor of California](https://www.gov.ca.gov/2026/09/18/governor-newsom-issues-executive-order-to-accelerate-independent-oversight-and-advance-the-creation-of-an-ai-kill-switch/) — "Governor Newsom issues executive order to accelerate independent oversight and advance the creation of an AI kill switch." - [Wall Street Journal](https://www.wsj.com/tech/ai/californias-newsom-issues-executive-order-to-weigh-ai-oversight-including-kill-switch-3a98040f) — "California's Newsom Issues Executive Order to Weigh AI Oversight, Including 'Kill Switch.'" - [Bloomberg Law](https://news.bloomberglaw.com/business-and-practice/newsom-signs-order-requiring-ai-labs-to-develop-kill-switch) — "Newsom Signs Order Requiring AI Labs Develop 'Kill Switch'" (auditor-qualification deadline moved to May 2027).