Claude IP allowlist: your API key now works only from your network

On November 5, 2026, Anthropic released the Claude IP allowlist in beta: a list of allowed IP addresses for a Claude organization. According to the Claude Release Notes, requests from addresses outside the list are rejected, which sharply limits what anyone can do with a leaked API key.

How does the Claude IP allowlist work?

An admin or owner of the organization enters the address ranges allowed to connect to Claude in the settings. For every covered request, Anthropic checks the sender’s public IP address against the list and rejects anything that isn’t on it (per the documentation).

In practice:

1. The API returns a 403 permission_error with the message “Your IP address isn’t in this organization’s allowed range.”
2. A person connecting from a network that isn’t allowed sees a page saying “Your current network can’t access this organization”. Admins get blocked too, unless we tick the option that keeps their access to the allowlist settings from any address.
3. The list holds up to 1,000 entries: single IPv4 and IPv6 addresses or CIDR ranges, no wider than /8 for IPv4 and /32 for IPv6. Public addresses only, not private ranges like 10.0.0.0/8.
4. Changes take effect within 15 minutes of saving.

What exactly does the list cover?

In Claude Platform (API) organizations, the block applies to the Claude Console and to the Messages, Message Batches, Token Counting, Models, Files and Skills endpoints, as well as the Claude Managed Agents endpoints and the Admin and Compliance APIs.

Interestingly, Claude Enterprise organizations get a separate list. There it protects not only against a leaked key but also against a stolen login: it covers Claude in the browser, the desktop and mobile apps, Claude Code (when signed in with a Claude account), Claude in Chrome, the Office add-ins, Cowork and Remote Tools. The owner and primary owner can set it up.

What should we watch out for?

First, what the list doesn’t cover. In Claude Platform it doesn’t apply to Amazon Bedrock, Claude Platform on AWS, Claude Models on Google Cloud or Microsoft Foundry, and in Managed Agents it protects only the endpoints, not the agents’ network traffic. In Enterprise, Claude Design, Claude in Slack and Claude Academy stay outside the list. A task Claude is already working on in a chat can also keep running for up to 24 hours after the user is blocked.

Second, the risk of locking ourselves out. When we save, the settings check only our current address, not our servers, CI or teammates. A phone on mobile data or a laptop without VPN will be cut off (before turning it on, list every network, including the ones that run jobs once a week).

Third, the list limits where, not when. Anthropic recommends also giving keys a short expiry, or moving to Workload Identity Federation and dropping static keys altogether.

What else shipped on November 5?

The same Release Notes add a beta Admin API call to sign a Claude Enterprise member out on all devices (POST /v1/organizations/users/{user_id}/sign_out with the admin-user-sign-out-2026-10-08 header). According to the documentation, it ends sessions, trusted devices and OAuth tokens issued for the organization, but API keys keep working, as do that person’s Claude Code cloud sessions and routines.

Is this anything new in AI?

Leaked API keys, as we know, aren’t theoretical: I wrote about it with the OpenCode leak, and about keeping agents in check in my guide to AI agent security. An IP list doesn’t make the problem go away, but a key found in a public repo stops working on someone else’s server. And if we track API spending, the breakdown of Claude API costs by key will come in handy too.

FAQ

Is the Claude IP allowlist paid?

The documentation doesn’t mention any extra fee. The feature is in beta under your agreement with Anthropic (e.g. the Commercial Terms of Service), and its settings and behavior may change during the beta.

Where do I turn on the IP allowlist in Claude?

In Claude Platform, on the Identity and access page in the Claude Console; in Claude Enterprise, in claude.ai under Organization settings > Organization and access. Turn on the IP allowlist toggle, add the ranges and save.

What if we lock ourselves out?

Connect from a network on the list, ask another admin or owner to add your address, or, if you ticked the option that allows settings access from any IP, add your current address yourself. As a last resort, Anthropic support can turn the list off.