---
title: California proposes kill switch for AI models
url: https://www.dataloco.com/en/california-proposes-kill-switch-for-ai-models
published: 2026-09-23T16:21:27+00:00
language: en
section: AI
source: https://www.punto-informatico.it/california-kill-switch-fermare-modelli-ai/
organizations: california, governor, dario amodei, sam altman, satya nadella
publisher: Dataloco
---

# California proposes kill switch for AI models

California's Governor announced the signing of an executive order that accelerates the implementation of new forms of third-party oversight concerning risks related to AI systems. This initiative, influenced by industry leaders, includes a proposal for the development of a kill switch to deactivate AI models in case of danger.

The Governor has previously signed legislation establishing a framework for independent organizations to assess AI systems in terms of safety and risk. Another law creates a state registry for AI auditors responsible for evaluating the compliance of AI systems with state regulations, setting standards for independence, transparency, and integrity.

In light of recent incidents and the inertia of the federal government, the Governor's executive order directs the Government Operations Agency to expedite the implementation of these laws. It also calls for the creation of a world-class group of experts to provide proposals for enhancing AI safety regulations within two months.

The proposals include at least four obligations. The kill switch should be activated in emergencies when a company loses control over its AI model or agent. However, experts warn that such a switch could be ineffective since AI systems are distributed across various servers, making it insufficient to simply disconnect one computer.

Additionally, a system equipped with a kill switch may be more vulnerable to cyber attacks. There is also a greater risk that a superintelligence could replicate itself in other infrastructures, leaving instructions for a network of AI agents to find ways to dismantle the kill switch. Thus, once control of the AI is lost, humans may have no chance of shutting it down.

## This story in other languages

- [Norsk](https://www.dataloco.com/no/california-onsker-en-kill-switch-for-a-stoppe-ai-modeller)
- [Русский](https://www.dataloco.com/ru/kaliforniia-xocet-vnedrit-perekliucatel-otkliuceniia-dlia-ostanovki-modelei-ii)
- [Nederlands](https://www.dataloco.com/nl/de-california-wil-een-kill-switch-voor-ai-modellen)
- [Français](https://www.dataloco.com/fr/la-californie-propose-un-kill-switch-pour-arreter-les-modeles-dia)
- [Português (Brasil)](https://www.dataloco.com/pt-br/a-california-propoe-um-kill-switch-para-sistemas-de-ia)