Playbooks · Teams tasks
Lock down Claude: IP allowlist and sign-out
Written November 6, 2026
Claude Platform adds beta IP allowlisting and an admin API that signs a member out everywhere. What each control covers — and its limits.
Two new controls, both in beta
The Claude Platform release notes for November 5, 2026 add two admin controls. The first is an IP allowlist for your organization, set up in the Claude Console. The second is an Admin API call that signs a member of a Claude Enterprise organization out of Claude everywhere. Both are in beta — neither is generally available, and beta behaviour can change.
IP allowlist: where a key works, not for how long
An organization admin or owner sets up a list of IP ranges in settings, and the list covers the whole organization. For each covered request, Anthropic compares the request's public IP address with the list and refuses it if it is not on the list. A leaked API key becomes much harder to use from anywhere else, which can prevent data leaks and reduce the chance of being charged for abusive use. But an allowlist limits where a key works, not for how long. Anthropic's own docs say to also set a short key expiration, or move to Workload Identity Federation to stop using static keys — safer than allowlisting alone.
Entries can be IPv4 or IPv6 addresses, or CIDR ranges no broader than /8 for IPv4 or /32 for IPv6. The list holds up to 1,000 entries per organization. List public addresses only — the addresses your traffic leaves your network from, not private ranges such as 10.0.0.0/8. If you add a shared proxy address, note that an attacker with a leaked key or a stolen sign-in could potentially use it through the proxy.
What the allowlist covers
For Claude Platform organizations, the allowlist covers the Claude Console UI, the Messages, Message Batches, Token Counting, Models, Files, and Skills API endpoints, the Claude Managed Agents endpoints (Agents, Sessions, Environments, Vaults, and Memory stores), and the Admin API and Compliance API. It does not apply to Claude in Amazon Bedrock, Claude Platform on AWS, Claude Models on Google Cloud, or Claude in Microsoft Foundry. It also covers requests to the Managed Agents endpoints, not the agents themselves, and does not apply to networking inside the Managed Agents runtime.
For Claude Enterprise organizations, the allowlist covers Claude in a browser, the desktop apps, and the mobile apps; Claude Code when users sign in with their Claude account; Claude in Chrome, the Office add-ins, the desktop file agent, and Remote Tools; and the Admin API and Compliance API. It does not cover Claude Design, Claude in Slack, or Claude Academy. A blocked user sees a page titled "Your current network can't access this organization"; in Claude Code, a blocked user sees a 403 API error. Blocked users can still sign in and use other organizations or a personal account. Phones on mobile data, home wireless, and laptops off the VPN are blocked until they connect through a listed network. And if Claude is working on a task in a chat when a user is blocked, the task can continue for up to 24 hours — the user cannot open or control that chat from outside the list.
Do not lock yourself out
List every network first: data centers, CI systems, offices, VPNs, proxies, and the workers of any self-hosted sandboxes, with their public IPv4 and IPv6 addresses. Home and mobile addresses can change without notice. Anthropic recommends selecting the checkbox that lets admins or owners reach the allowlist settings page from any IP address, so an admin can fix the list from anywhere. The downside is that an attacker who compromises an admin's or owner's session could update the allowlist from a remote location. Changes can take up to 15 minutes to apply, so watch for 403 permission_error responses — including from jobs that run only daily or weekly — and add any address you missed.
Sign out everywhere: a session kill, not a revocation
POST /v1/organizations/users/{user_id}/sign_out signs a member out of Claude on every browser and device, without removing them from the organization. It takes no body — just the member's ID — and needs the write:members scope. Anthropic's docs name the use cases plainly: a member's device is lost or stolen, a security incident is under way, or you are investigating an account before deciding whether to remove the member. An owner can also use the Sign out everywhere action on the Members page of claude.ai organization settings instead; an owner cannot use it on their own account.
The sign-out ends all sign-in sessions for the person's whole Claude account, on every browser and device; every trusted device they registered for sensitive actions; OAuth tokens issued for this organization, held by whichever app — Claude Code or a third-party app — which must sign in again; the record of browsers where the person chose to skip two-step verification, so the next sign-in asks for the second step again; and push-notification registrations. A token almost always stops working at once, and within an hour at most.
What it does not end matters more. The member keeps their seat and role and can sign in again right away. To keep someone out, remove them or, if your organization requires single sign-on, disable their account in your identity provider. The request does not end API keys — they keep working, including Admin API keys the person created. It does not end OAuth tokens for other organizations, or tokens issued to the person's account rather than to an organization. It does not end Claude Code cloud sessions and routines: sessions running on Anthropic's infrastructure keep running, get new credentials without the person signing in again, and continue; scheduled routines keep starting new sessions. Computers linked to the desktop file agent stay linked. And a sign-in already under way can finish during the request, or up to 5 minutes after it.
What neither control does
Neither control replaces removing the member. The allowlist narrows where a stolen key or sign-in can be used from; sign-out ends live sessions. If someone should no longer have access, remove them from the organization — and if your identity provider manages membership, remove them there, because automatic provisioning can add a removed person back at their next sign-in if their identity provider account stays active. Removing a member returns any purchased seat they occupied to the organization's pool.
Limits
- Both controls are in beta. The release notes describe them as beta, and the sign-out endpoint is marked beta in the Admin API docs.
- No control on this page was tried in a console. This page follows the release notes and the linked docs; no allowlist was set up and no member was signed out while writing it.
- The allowlist and sign-out behave differently across organization types. The endpoints and apps listed above are split between Claude Platform (Claude Console) and Claude Enterprise (claude.ai) organizations — check which type yours is before following either set of steps.
Related playbooks
- Fulcrum's MCP sign-in goes OAuth (alpha)
- Claude Meetings, Sheets and Whiteboards land
- Goldman Sachs limits Claude Code to engineers
- Infor's connector lets Claude act, with approval
Sources
- Claude Platform release notes — Anthropic
- IP allowlisting — Claude Platform Docs
- User management — Claude Platform Docs
Back to: Small-team playbooks hub · Also: Verify what managed settings auto-apply after 2.1.316 · Stop an agent running up a bill · Audit a repository's git config before any agent opens it