---
title: What Anthropic Admin API access does an AI governance tool need, and what can it not see?
url: https://opencontroller.lyzr.com/blog/anthropic-admin-api-access-for-ai-governance-scanning
published: 2026-11-16
updated: 2026-11-16
author: Lyzr
---

# What Anthropic Admin API access does an AI governance tool need, and what can it not see?

**An AI governance scan of Anthropic needs an Admin API credential: an `sk-ant-admin` key, an `org:admin` OAuth token, or an unscoped personal or service account key. With it, the scan can list members and their roles, workspaces, workspace members and API keys. What it often cannot see is group membership, because group and custom-role reads are Enterprise-only.**

That last point came up on a recent buyer call. The team had connected Anthropic, saw every key and user, and then asked why no RBAC groups appeared. The answer sits in the API, not the scanner. This guide covers what to grant, what comes back, and how to fill the gaps.

For the wider picture across vendors, see [how to govern AI agents across platforms](/blog/how-to-govern-ai-agents-across-platforms).

## Which credential should the governance tool use?

Anthropic's Admin API accepts three kinds of credential. Pick the one that matches how you run integrations.

| Credential | Header | Who can create it | Notes |
|---|---|---|---|
| Admin API key (`sk-ant-admin...`) | `x-api-key` | Members with the admin role; owners and primary owners | Simplest option for a scanner |
| OAuth bearer token with `org:admin` scope | `authorization: Bearer` | Admin, owner or primary owner | Required for service account and federation endpoints |
| Personal or service account key not scoped to a workspace | `x-api-key` | Depends on the linked account | Same permissions as the linked account |

Two limits matter before you start:

- The Admin API is available to **Claude Console** and **Claude Enterprise** organizations. It is not available on the **claude.ai Team plan** or to individual Console accounts.
- **Claude Enterprise** (claude.ai) organizations use a scoped key created in claude.ai. Only the members and invites endpoints apply to them.

A practical rule: create the admin credential under a named integration owner, store it in your secrets manager, and record it in your agent inventory like any other privileged credential.

## What does the Admin API return?

With an admin credential, a scan can read:

1. **Organization members**, each with an organization role.
2. **Pending, accepted and expired invites.**
3. **Workspaces** and **workspace members** with their workspace roles.
4. **API keys**, filterable by `workspace_id` and `status`, with metadata.
5. **Organization info** via `GET /v1/organizations/me`.
6. **Usage and cost** through the Usage and Cost Analytics API (the older Usage and Cost API is legacy).

API key metadata is the most useful part for governance. Each key carries:

- `expires_at`: `null` means the key never expires.
- `principal`: the user or service account the key acts as, or `null` for a workspace key.
- `scope`: bound to one workspace, or organization-wide.

A key with `expires_at = null`, `principal = null` and organization-wide scope is the first thing to review. It never expires and has no named owner.

## What are the Anthropic organization roles?

The Admin API documents five organization roles:

| Role | What it allows |
|---|---|
| `user` | Use the playground |
| `claude_code_user` | Playground and Claude Code |
| `developer` | Playground and manage API keys |
| `billing` | Playground and manage billing details |
| `admin` | All of the above, plus manage users |

Owners and primary owners hold admin permissions and can also manage admins. Workspace roles include `workspace_developer` and `workspace_admin`.

For a governance review, the `developer` role deserves attention. Anyone with it can create keys, and keys are how agents reach models.

## Why are groups missing from the scan?

The Admin API exposes **group and custom-role reads only on Enterprise**. On a Console organization without them, a scanner can only report:

- each member's single organization role, and
- their role in each workspace.

So "who in the payments team can create keys" cannot be answered from Anthropic alone. You have three options:

1. **Join with your identity provider.** Match member emails to IdP groups and attach the group as an owner attribute in your inventory.
2. **Use workspaces as the grouping.** Create one workspace per team or agent family, so workspace membership stands in for group membership.
3. **Use the Compliance API on Enterprise.** Anthropic points to it for organizations, users, roles, groups and settings data.

Whichever you choose, document it. An auditor will ask how you know who owns each key.

## What checklist should a governance scan run on Anthropic?

1. List every API key. Flag keys with no expiry.
2. Flag keys whose `principal` is a person who has left, or is `null` with organization scope.
3. List members with `developer`, `admin` or owner rights. Compare with the approved list.
4. Check each workspace has a named owner and a spend limit.
5. Pull usage by key and workspace. Flag keys with spend but no registered agent.
6. Repeat on a schedule, not once.

The [guide to finding every Anthropic API key](/blog/find-anthropic-api-keys-across-workspaces) covers step 1 in more depth. [Workspace spend limits](/blog/anthropic-workspaces-for-agent-spend-limits) cover step 4.

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys, and connects to Claude Enterprise and OpenAI Enterprise workspaces. It also discovers and manages skills in Anthropic Enterprise. Discovered keys and agents land in a governed registry with owners and versions. Posture then flags risky permissions and configuration, such as keys without an owner. Access to Opencontroller itself uses OIDC SSO, so roles map from your identity provider. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Can a workspace-scoped key list all API keys?**
No. Organization-wide listing needs an Admin API credential. A workspace key only sees its own workspace.

**Does the Admin API create API keys?**
The documentation covers listing keys and updating their name or status (active or inactive). It does not document a create call, so plan for key creation in the Console.

**What should we do first with the scan results?**
Deactivate or reassign keys that never expire and have no named owner, then require every remaining key to map to a registered agent.

## Key facts

Anthropic Admin API keys start with sk-ant-admin and can only be provisioned by organization members with the admin role (owners and primary owners also qualify). The Admin API is available to Claude Console and Claude Enterprise organizations, not the claude.ai Team plan or individual Console accounts. Organization roles are user, claude_code_user, developer, billing and admin. Group and custom-role reads are Enterprise-only. Claude Enterprise organizations use a scoped key created in claude.ai, and only the members and invites endpoints apply to them. Opencontroller discovers agents, models and API keys and connects to Claude Enterprise workspaces. Opencontroller pricing starts from $250K per deployment per year.

## Related

- [How do you find every Anthropic API key across workspaces?](https://opencontroller.lyzr.com/blog/find-anthropic-api-keys-across-workspaces.md)
- [How do you use Anthropic Console workspaces to cap agent spend?](https://opencontroller.lyzr.com/blog/anthropic-workspaces-for-agent-spend-limits.md)
- [How do you track Claude API spend per agent with Anthropic's usage and cost APIs?](https://opencontroller.lyzr.com/blog/anthropic-usage-and-cost-api-for-agent-spend.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
