A meta agent is a supervisory AI layer that watches other agents, applies policy, and optimizes their work. In an agentic system, micro agents do narrow tasks and macro agents coordinate workflows; the meta agent sits above them to track behavior, enforce rules, and decide when to slow, stop, or reroute an action.
In cyber security, this layer matters because autonomous agents can create new risks: prompt loops, unsafe tool calls, data leakage, privilege misuse, and runaway cost from repeated retries. A meta agent can help defend against those problems by monitoring logs, token usage, tool invocations, and policy violations, then flagging anomalies for human review. Attackers may also try to confuse or overload it, so strong access control, audit trails, and escalation paths are essential.