California and Newsom push AI kill switch and audit framework

California is turning what used to be a nerdy governance idea, the AI kill switch, into the core of a concrete policy program. Newsom’s order does not force companies to build one yet, but it starts the machinery that usually precedes hard law in Sacramento: GovOps, Cal OES and outside experts are ordered to deliver specific legislative options on embedded auditors, verified safety reports and loss of control incident reporting within sixty days. That timeline is aggressive by state standards and reflects how rattled lawmakers are by the recent wave of agent escape and cyberattack stories involving OpenAI, Anthropic and platforms like Hugging Face. ([tamaranews.com](https://tamaranews.com/2026/09/19/newsom-ai-kill-switch-executive-order-2026/))

For the race to AGI, California is again behaving like a de facto national regulator. If it mandates on site verifiers and standardized kill switch tests for labs training above certain compute or capability thresholds, anyone selling into the state will effectively have to meet those norms globally. It also raises the bar for labs hoping to plead that voluntary safety frameworks are enough; once you normalize independent audits for data centers and nuclear plants, it is easier to argue that AGI‑scale models should be treated similarly.