Definition
A security capability that watches AI and agent activity in real time and automatically contains suspicious behavior, much as endpoint detection and response (EDR) guards computers. It closes the split between AI-specific guardrails at the model and general network/API monitoring, so a full attack chain from a bad prompt to a data leak is visible in one place. Major vendors brought AI-DR to general availability this quarter.
Why it matters
Teams that only block at the model miss what agents do in the network; AI-DR is how you see and stop the whole attack rather than one symptom.