A control that lets people quickly stop an AI model or agent, cutting off its output or access if it starts doing something harmful.
AI Drafted with AI help · Reviewed by Keshav Aggarwal · Updated
In simple words
A treadmill has a red safety clip: if you slip, the clip pulls out and the belt stops at once. An AI kill switch is that safety clip for AI.
If an AI helper starts doing the wrong thing, a person can press stop and cut it off before it does more.
The technical side
An AI kill switch is a way to halt an AI system fast, for example by blocking its access to the model, its tools or the network, rather than waiting for a fix. It matters most for AI agents that act on their own, and it does not undo actions an agent has already taken.
On 3 November 2026 CrowdStrike launched an AI Kill Switch in its Falcon Guardian platform: security staff can stop a rogue internal agent, stop it relaunching and remove its network access, by having AI gateways cut it off from the models it relies on, though a person still has to decide to use it. On 11 November 2026 OpenAI described Control Room, a system across its research infrastructure that can cut off model output from a research job within 35 seconds of receiving a stop instruction, and that denies model requests if Control Room itself goes down. More on our AI news feed.
AI Kill Switch kya hai?
AI kill switch ek emergency button jaisa hai: agar AI agent galat kaam karne lage, to insaan use turant rok sakta hai aur uska access kaat sakta hai. Nov 2026 mein CrowdStrike aur OpenAI ne aise systems ke baare mein bataya. Par jo kaam AI pehle hi kar chuka hai, wo ye button waapas nahi le sakta.
Related terms
- AI AgentAIAn AI system that can plan steps and use tools on its own, such as searching, clicking or sending messages, to finish a task.
- Human in the LoopAIAn AI set-up where a person checks or approves the AI’s work at the important steps, so it never makes big decisions alone.
- Red TeamingAITesting an AI by attacking it on purpose, trying to make it break rules or cause harm, so the problems are fixed before the public uses it.
- Prompt InjectionAIAn attack where hidden or sneaky text tricks an AI model into ignoring its instructions and doing something it should not.