What the new call does, and what it does not
On November 5, 2026 Anthropic announced that you can sign a member of a Claude Enterprise (claude.ai) organization out of Claude everywhere with the Admin API, in beta. The member stays in the organization and can sign in again.1 The endpoint is POST /v1/organizations/users/{user_id}/sign_out, listed beside list, update and remove on the members resource.2
Three limits matter for a runbook:
- It is a session action, not a removal. A person who still has a valid identity provider login can come back in.
- It is documented on the Claude Enterprise user-management page. I did not confirm that Claude Console (Claude Platform) organizations get it, so check your own organization before wiring it in.
- It is beta. Anthropic says beta features may have breaking changes with notice, be deprecated or removed.3 Wrap the call so a failure pages a human instead of silently skipping a step.
Anthropic's release note says the user-management page gives the beta header and what the request does not end.1 Read that section before you ship; this brief does not reproduce the header value.
Where it sits in the leaver runbook
Order matters, because the member can sign in again.
- Disable the account in your identity provider. Otherwise the next sign-in recreates the session you just ended.
- Look up the member id.
GET /v1/organizations/userstakes an email filter, and the match tolerates common variants of an address.2 - Sign the member out everywhere (beta). Keep this for people who stay in the organization, such as a role change or a suspected compromise, and as a fast first step for a leaver.
- Remove the member with
DELETE /v1/organizations/users/{user_id}for a true leaver. On plans with a finite seat pool, removal returns the seat.2 - Work through the credentials below.
- Verify and log (last section).
# Placeholders: ADMIN_KEY needs write:members; the beta header value comes from
# Anthropic's user-management page, not from this post.
curl -sS -X POST \
"https://api.anthropic.com/v1/organizations/users/$USER_ID/sign_out" \
-H "x-api-key: $ADMIN_KEY" \
-H "anthropic-version: 2026-10-21" \
-H "anthropic-beta: <beta-header-from-docs>" \
-o signout-response.json -w "%{http_code}\n"Member POST and DELETE endpoints need an Admin API key with the write:members scope, and Admin API endpoints share a limit of 100 requests per minute per organization.2 A bulk offboarding of a department should throttle itself.
Owners are the exception. Members holding owner, membership_admin or primary_owner cannot be modified or removed through the members API; those roles are assigned in claude.ai organization settings.2 Put a manual step in the runbook for them.
What a session sign-out leaves live
Ending sessions does not end the things a person created. Each of these is a separate resource with its own step.
| Still live after sign-out | Why | What to do |
|---|---|---|
| Console API keys | The Admin API manages API keys as their own resource, separate from members4 | List keys, find the ones the leaver created or used, disable or rotate them, and redeploy the applications that held them |
| Personal or service account keys | A personal key not scoped to a workspace has the same permissions as the linked account, and works against the Admin API4 | Confirm removal or deactivation also kills these; test a call with a retired key |
| Admin API keys and OAuth tokens | Admin keys come from admins; org:admin OAuth tokens are issued to admin, owner and primary owner roles4 | Rotate any Admin key the leaver created or could read |
| Workspace membership | Workspace members are managed through their own endpoints4 | List the workspaces and remove the user where removal from the organization is not enough |
| Gateway keys | They live in your gateway, not at Anthropic | Revoke or rotate them there |
| Copies of a key | In CI variables, notebooks, local .env files | Treat any key the person could read as exposed; see the leaked Anthropic key guide |
I could not confirm from the pages read whether removal also ends sessions or disables personal keys, so verify each rather than assuming.
Shared workspaces and gateway keys
A shared workspace keeps running after its creator leaves. List its members and its keys, reassign an owner, and rotate keys the leaver could see. The same holds for gateway keys: an application holding a gateway key keeps working after its author is gone, which is the point of the key, but someone must own it.
If you run INFRO, the roster and keys sit in one place. SCIM deactivation of members is available from the Business plan, and a revoked key returns 401, which gives you a test to save. Key and membership changes are recorded in the audit log with actor and timestamp.5 Key limits and rotation with an overlap window are shipped, so a rotated key can keep working briefly while you deploy its replacement.
Evidence to log for each leaver
An auditor wants to see that the action happened, who did it and that it worked. Per leaver, store:
- Member id, email, role and the HR ticket id.
- UTC time of each step: identity provider disable, sign-out, removal.
- The actor, meaning the Admin API key id or the person who ran it, never the key value.
- The raw HTTP status and response body of each call.
- A follow-up read, such as the member lookup returning nothing, saved with its time.
- A list of API keys, workspace memberships and gateway keys reviewed, each with its disposition (revoked, rotated, reassigned, none found).
- For each revoked key, one saved failing call. INFRO's access review post makes the same point: calling a key you revoked and saving the 401 proves it took effect.5
Then reconcile quarterly: the provider member list, your HR leavers list and the gateway roster, as in the access review checklist. A leaver who still appears in any of the three is a finding. Anthropic also lists a Compliance API and Activity Feed in its docs; I did not read those pages, so check whether they add session-level evidence for your plan.
