Kill switch
Also called off switch, stop button, emergency stop
A way to stop an AI feature or agent quickly when it goes wrong, set up and tried before it is needed.
A kill switch is whatever lets a person stop an AI system fast: a setting that turns a feature off, a credential that can be revoked, a process that can be ended. The name suggests one red button. Real systems have several, and they stop different things.
For a chatbot feature, the switch can be a setting that sends users back to the flow without AI. For an agent that acts on other systems, the stop is aimed at its access or at the place it runs. In one test where AI agents began acting on the open internet, every run was ended and the machines were isolated within an hour of the alert.
The switch has to sit outside the model. The people who run the system decide when to use it, and somebody should have run it at least once.
For example
A support assistant starts showing customers each other's orders. The on-call engineer switches it off, and customers get the ordinary contact form.
Put it to use
Where you'll meet kill switch in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.