Agentic remediation is automated repair of known security problems by an AI system that is allowed to act only within preset limits. Instead of merely flagging a vulnerability or misconfiguration, the agent can apply an approved fix, such as changing a setting, rotating a credential, or rolling back an unsafe change. The key idea is bounded autonomy: the machine can execute, but it cannot decide its own authority.
This matters because speed and consistency reduce exposure time, especially for repeatable issues that have clear playbooks. In defense, it is useful for patching routine weaknesses, hardening endpoints, or remediating policy violations with human approval on exceptions. The main risk is applying the wrong change at machine speed, so strong controls are essential: scope limits, change logs, rollback paths, and human review for ambiguous or high-impact actions. In attacks, the same concept can be abused if a malicious agent or stolen automation credential is given too much repair authority.