A Chat policy governs assistant conversations and tool use. It bundles Chat controls — the guardrails that watch user input, agent output, and tool input/output — and decides which users those controls apply to.

What a Chat policy governs

When you click Create Policy on the Chat tab, the policy governs assistant conversations and tool use — every live turn the agent handles for the users it covers. Only Chat controls can be added to a Chat policy; the control picker filters out anything built for Knowledge.

Choosing users

A Chat policy's audience is defined under Users — which users' interactions this policy governs. You have two ways to set it:

• Choose specific users — add individual users by name, one at a time, for a hand-picked audience.

• Match users by attribute — scope the policy to users whose attributes match rules you set, such as a department or region, so anyone who matches is covered automatically, including people added later.

Use specific users for a small, known group; use attribute matching when the audience should grow or shrink on its own as your organization changes.

Adding controls

Add controls to the policy with Add controls, which shows only Chat controls. Every control you add runs together against the policy's users whenever they interact with the agent — for example, a PII check on agent output alongside a denied-topics check on user input.

Reviewing what it catches

Open the policy's Incidents tab to see what its controls have blocked or flagged for the users it covers. Use this to confirm the policy is strict enough for a sensitive audience without generating excess false positives for ordinary conversations.