Claude Adds IP Allowlists and Enterprise Sign-Out API: Coverage and Limits

Anthropic's November 5 platform release notes introduce two security controls in beta: an organization IP allowlist configured in the Claude Console, and an Admin API operation that signs a Claude Enterprise member out across browsers and devices.

The allowlist accepts up to 1,000 IPv4 or IPv6 addresses and CIDR ranges. It rejects covered requests from other addresses, limiting where someone can use a leaked API key. The Enterprise operation ends sign-in sessions, trusted-device registrations and OAuth tokens issued for that organization. Membership remains intact, and the person can sign in again immediately.

These controls are useful additions to an incident-response plan. Teams should decide which access they need to stop before choosing an action.

Check where the IP rule applies

The IP allowlisting guide covers the Console, specified Claude API and Managed Agents endpoints, and organization-scoped Admin and Compliance API credentials. The rule checks incoming requests. It does not restrict networking inside a Managed Agents runtime.

It also does not apply to Claude in Amazon Bedrock, Claude Platform on AWS, Claude Models on Google Cloud or Claude in Microsoft Foundry. Teams using those routes need to inspect their respective access controls rather than assuming a Console setting protects every deployment.

Use public egress addresses, including the routes used by CI and proxies. Private addresses are not suitable. Changes can take up to 15 minutes to apply; blocked API calls return a 403 permission error.

Admins can optionally retain access to the allowlist settings from any address. That offers a recovery path after a mistake, but an attacker with a compromised admin session could also change the list remotely.

Treat rollout as a network change. Identify an owner, record the intended addresses and schedule a test covering occasional jobs as well as the main application. A successful request from your laptop says little about a weekly export running elsewhere.

Signing out has a wider personal reach

The new endpoint is POST /v1/organizations/users/{user_id}/sign_out. Every request needs the admin-user-sign-out-2026-10-08 beta header.

Anthropic's Enterprise user-management documentation requires an organization-specific Admin API key with the write:members scope. Organizations without beta enrollment can receive an error. The primary owner cannot be signed out through this endpoint.

Session revocation reaches the person's whole Claude account, including other organizations and an individual plan. OAuth revocation is narrower: it covers tokens for the requesting organization. Those tokens normally stop working immediately, but can take up to an hour.

That difference matters when writing an incident playbook. An administrator should know that the action can interrupt the person's other signed-in Claude sessions, while still leaving some credentials usable.

Check what continues running

The documentation lists important exclusions. API keys remain active, including Admin API keys. Claude Code cloud sessions and routines can keep running, and Cowork computers remain linked. Data already stored on a device is not deleted. A sign-in already underway can also finish during the request or for up to five minutes afterward.

For a suspected compromise, inventory the affected access separately: the user session, API credentials and active jobs. Record which action addresses each one and how the team will confirm the result. Avoid using a successful sign-out response as the sole completion check.

The same discipline applies to the new IP control. Test both an approved path and a deliberately unapproved path with non-sensitive requests. Keep credential rotation and the investigation of prior activity in the response plan. Restricting where future requests originate does not explain what happened before the restriction.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help technical founders turn AI infrastructure and developer-tool updates into clear technical writing and product stories.

Book a 30-minute call.