Definition
A security product category that watches AI agents while they run, learns what each agent is allowed to do, and stops malicious or unintended actions in real time. It combines discovery, policy enforcement, and behavioural monitoring for agents specifically — the way endpoint detection responds to malware on computers.
Why it matters
Agents hold real credentials and can act at machine speed, so a rogue agent can cause damage before anyone notices. ADR is the runtime safety net that detects and halts those actions mid-flight.