---
title: How do you map identity provider groups to admin and viewer roles in an AI governance tool?
url: https://opencontroller.lyzr.com/blog/map-idp-groups-to-roles-in-ai-governance-tools
published: 2026-11-13
updated: 2026-11-13
author: Lyzr
---

# How do you map identity provider groups to admin and viewer roles in an AI governance tool?

**Create one identity provider group per console role, map each group to exactly one role by its immutable group ID, and make "no access" the default for anyone unmapped. Then test the mapping after every sync with one known user per role, because most role errors come from missing group claims, renamed groups or nested groups that never expand.**

An AI governance console controls which agents run, which models they reach and which policies apply. Who holds admin there matters as much as who holds admin in your cloud accounts. Yet role mapping is often set up once during a pilot and never tested again.

## Which roles does an AI governance tool need?

Keep the set small and separate the people who change things from the people who check them.

| Role | Can do | Should not do | Typical IdP group |
|---|---|---|---|
| Platform admin | Configure SSO, connectors, integrations | Approve their own policy exceptions | ai-gov-platform-admins |
| Policy admin | Write and change policies and guardrails | Approve exceptions to their own policies | ai-gov-policy-admins |
| Approver | Approve agent versions and exceptions | Edit agents or policies | ai-gov-approvers |
| Operator | Quarantine, block, investigate alerts | Change policies or connectors | ai-gov-operators |
| Auditor | Read everything, export evidence | Change anything | ai-gov-auditors |
| Viewer | Read dashboards for their own agents | Read other teams' data or evidence | ai-gov-viewers |

Two to five people in platform admin is usually enough. If more than that need it, some of them probably need policy admin or operator instead. For the approver role, apply the maker-checker rule in [separating who builds an agent from who approves it](/blog/segregation-of-duties-for-ai-agent-approvals).

## Why does an admin show up as a viewer?

When a known admin lands with the wrong role, check these causes in order:

1. **The group claim was not in the token.** Microsoft Entra ID, for example, stops listing groups in the token when a user belongs to more than 200 groups in a JWT or 150 in a SAML assertion, and sends an overage indicator instead. The app then sees no groups and falls back to the default role. Filter the claim to groups assigned to the application.
2. **The mapping used a group name.** Someone renamed the group, and the mapping no longer matches. Map by immutable group ID.
3. **Nested groups were not expanded.** The admin is in a group inside the mapped group. Many provisioning and claim setups do not flatten nesting, so assign users directly to the role group.
4. **The user is outside provisioning scope.** SCIM only provisions users and groups assigned to the application.
5. **The role was set at first login and never refreshed.** Some apps read groups only on account creation. Check whether roles update on each sign-in or each SCIM sync.

SCIM, defined in RFC 7644, carries users and group memberships to the app. The sign-in token carries claims at login. If the two disagree, decide which source wins and document it.

## How should the mapping be configured?

Follow a short checklist:

1. **One group per role.** Do not reuse broad groups such as "Engineering" for console roles.
2. **Map by group ID**, and record the group name alongside it for readability.
3. **Set the default role to no access**, or to viewer if every employee should see their own agents.
4. **Choose a precedence rule** for users in several groups, usually the highest privilege wins, and test it.
5. **Keep break-glass accounts outside the mapping**, in a vault, with alerts on use.
6. **Review group membership quarterly** in the same cycle as [access reviews for AI agents](/blog/access-reviews-for-ai-agents).

## What should you test after each sync?

Keep a test user in each role group and run five checks after every mapping change and on a regular schedule:

- Each test user lands with exactly the expected role.
- A user removed from a role group loses the role within one sync cycle.
- A user in two role groups gets the role your precedence rule says.
- An unmapped user gets no access or viewer, never admin.
- Every role change is recorded in the tool's audit log with the source group.

Record the results as evidence. Auditors will ask how you know the console's roles match the identity provider. The [identity pillar](/blog/ai-agent-identity-on-behalf-of-delegation) covers the agent side of the same question.

## How Opencontroller does this

Opencontroller uses OIDC SSO for console access, so sign-in, multi-factor policy and offboarding follow your identity provider. It is built on open standards, including OIDC, SAML and SCIM. To review role design for your own deployment, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Why does an admin show up as a viewer after SCIM sync?**
Usually the group claim was missing from the sign-in token, the mapping used a group name that changed, a nested group was not expanded, or the role was set at first login and never refreshed. Check the token claims first.

**Should anyone hold a role directly instead of through a group?**
Only break-glass accounts. Everyone else should get roles through IdP groups, so joiners, movers and leavers are handled automatically.

**How many platform admins should an AI governance tool have?**
Usually two to five. Most people who ask for admin need policy admin, operator or auditor instead.

## Key facts

Map roles in an AI governance tool from identity provider groups, never from individual assignments. A least-privilege set: platform admin, policy admin, approver, operator, auditor (read-only including evidence) and viewer. Use one IdP group per role, map by immutable group ID rather than display name, and make the default role for unmapped users "no access" or viewer. Common causes of an admin appearing as a viewer after sync: group claim missing from the token (Microsoft Entra ID emits an overage indicator instead of groups when a user is in more than 200 groups for JWT or 150 for SAML), mapping by group name after a rename, nested groups not expanded, the user not in provisioning scope, or role assigned at first login and never refreshed. Test after every sync with one known user per role. Opencontroller by Lyzr uses OIDC SSO; pricing starts from $250K per deployment per year.

## Related

- [Should AI governance tools allow local logins or SSO only?](https://opencontroller.lyzr.com/blog/sso-only-access-for-ai-governance-tools.md)
- [How do you run access reviews for AI agents?](https://opencontroller.lyzr.com/blog/access-reviews-for-ai-agents.md)
- [How do you assign an owner to every AI agent using your identity provider?](https://opencontroller.lyzr.com/blog/assign-owners-to-ai-agents-with-your-identity-provider.md)
- [More on Agent identity and access](https://opencontroller.lyzr.com/blog/cluster/agent-identity-and-access)
