Skip to main content
All glossary terms

AI governance

Guardian Agents

Guardian agents are software agents or agent-like controls intended to observe or constrain the actions of another agent. The label is used for a design pattern, not a standardized technical category; implementations vary in authority, visibility, and ability to intervene.

Also known as agent guardrails.

Why it matters in practice

A monitoring agent can introduce its own failure modes, permissions, and dependencies, so its decisions need oversight just like the system it watches. Teams should define what signals it sees, what actions it can take, and how people can inspect or override those actions. A guardian design is not a complete safety guarantee.

Example

A review component checks proposed outbound actions against a policy and sends an uncertain case to a person instead of silently approving it.

Educational note

This definition is general information, not legal advice. It may omit facts that matter to a specific situation and does not establish or guarantee compliance.