Glossary · Updated July 2026

What is AI agent governance?

AI agent governance is the set of policies and controls that decide which AI agents may exist, what they are allowed to do, and how their actions are held to account — permissions, review, audit, and compliance applied to autonomous software. It is how an organization keeps a fleet of agents inside its risk, legal, and security boundaries.

Governance often starts top-down and abstract — risk registers, approval workflows, compliance frameworks — and its hardest problem is reaching the working loop where agents actually act. A policy that never touches the moment an agent runs a tool is theatre.

The enforcement point is review at the edge: risky changes checked before they land, with deterministic evidence rather than another model's opinion. Deterministic review is what survives an audit — the same input always yields the same, traceable verdict. In Vivari that role belongs to Guard, which gates changes with facts mined from your own git history and no LLM in the loop.

How it relates to agent management

Governance consumes the records that AI agent management produces. Management is the layer that makes governance enforceable — turning policy into permissions, gates, and an exportable trail.