Containment
Also called contain, incident containment
The first step of an incident response: stopping the harm from spreading before anyone starts on the cause.
When something goes wrong, the pull is to fix it. Containment comes first, because a fault that is still running keeps adding people to the list of those affected.
With AI there is a second reason. The quick fix within reach is an edit to the prompt, made while the assistant is still answering. That changes the system before anyone has saved what it was shown and what it did, so the evidence of the cause goes with it.
Containing an agent means cutting its access or ending its run. One team whose test agents began acting against real organisations had the incident contained within about an hour of finding it, and then spent days working out what had happened.
For example
An AI assistant is leaking customer records. The team switches it off and saves its records, and only then looks for the fault.
Put it to use
Where you'll meet containment 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.