Anthropic IP Allowlisting and Session Revocation Setup Guide
Configure Claude Console IP allowlists and Enterprise Admin API session sign-out to isolate leaked credentials and protect automated agent workflows.
- Anthropic launched organization IP allowlisting in beta on 2026-11-05, blocking leaked API tokens outside permitted CIDR blocks.
- Console and Enterprise admins can define up to 1,000 IPv4 and IPv6 entries spanning /8 to /32 boundaries.
- The 2026-11-05 release also introduced an Admin API endpoint allowing Enterprise admins to sign members out across all sessions.
- Allowlist rule propagations take up to 15 minutes to apply across covered endpoints, returning HTTP 403 errors on rejected requests.
Strengthening Claude API Security Perimeter
Leaked credentials remain a major operational risk for teams integrating LLM services into continuous integration workflows and automated backend servers. A rogue actor holding an API secret can trigger unauthorized inference calls, extract stored data, and run up significant charges before an engineering team spots unusual spikes. The release on 2026-11-05 addresses these vulnerabilities by introducing IP allowlisting alongside remote user sign-out capabilities across the Claude Platform and Claude Enterprise.
While key expiration limits the lifespan of a credential and Workload Identity Federation replaces long-lived tokens with federated tokens, network fencing establishes an immediate perimeter defense. By refusing requests originating from unlisted addresses, organizations constrain where a credential can execute, minimizing damage if keys are exposed. Developers handling production workloads can also review handling API errors and retries to gracefully account for network denials.
Covered Services and Protected Endpoints
The IP allowlist applies across both browser-based administrative surfaces and core Anthropic REST endpoints. In Claude Platform workspaces, enforcement blankets the Messages API, Message Batches, Files, Token Counting, Skills, and Models listings, whether requests authenticate via standard API keys or Workload Identity Federation tokens. Organizations managing enterprise governance through the Admin API and the Compliance API also inherit network isolation when utilizing single-organization keys, as previously discussed in the Anthropic Compliance API overview.
| Service Area | Covered Endpoints & Features | Excluded Services & Limits |
|---|---|---|
| Claude Platform | Messages, Message Batches, Files, Token Counting, Skills, Models, Claude Console UI | External runtime traffic from Managed Agents, Bedrock, AWS, Google Cloud, Microsoft Foundry |
| Managed Agents | Agents, Sessions, Environments, Vaults, and Memory stores management endpoints | In-agent external network requests and runtime execution inside sandboxes |
| Claude Enterprise | claude.ai browser, desktop/mobile apps, Claude Code, Chrome extension, Office add-ins, Cowork | Claude Design, Claude in Slack, Claude Academy, artifacts shared outside the organization |
| Administrative APIs | Admin API, Compliance API (when using organization-scoped credentials) | Keys that authenticate across multi-organization administrative contexts |
Fencing covers management APIs for Claude Managed Agents—including Environments, Sessions, Vaults, Memory stores, and Agents endpoints. However, the allowlist does not restrict external networking initiated within the agent runtime itself. Third-party cloud distributions, including Claude on Google Cloud, Claude in Amazon Bedrock, Claude Platform on AWS, and Claude in Microsoft Foundry, operate under their respective cloud provider networking configurations rather than Console allowlists.
Network Entry Rules and Configuration Workflow
To activate the security policy, an owner or administrator navigates to the Identity and access section of the Claude Console, or the Organization and access panel inside Enterprise settings. Enabling the toggle does not immediately disrupt traffic; configuration changes become active only after adding validated ranges and saving.
- CIDR Scope: Supports single addresses like 203.0.113.5 as well as subnets such as 203.0.113.0/24 or IPv6 blocks like 2001:db8::/32.
- Subnet Boundaries: IPv4 ranges cannot be wider than /8, while IPv6 subnets cannot exceed /32 breadth.
- Capacity: Up to 1,000 discrete address or CIDR entries can be configured per organization.
- Public Routing: Only external egress addresses are permitted; internal prefixes like 10.0.0.0/8 should never be entered.
Once saved, updates require up to 15 minutes to fully synchronize across Anthropic infrastructure. When an incoming API call originates from an unrecognized address, the Messages API terminates the TLS handshake with an HTTP 403 status code indicating a permission rejection because the address falls outside the organization’s approved ranges. Web and desktop clients encounter an access restriction warning.
Preventing Administrator Lockouts
Enabling a restrictive network perimeter without verified egress paths can inadvertently sever access for deployment servers and engineering teams. The Console automatically checks whether the active session’s public IP matches the submitted configuration before confirming, preventing accidental self-lockout during setup. However, this preliminary verification does not evaluate automated CI runners, remote developers, or scheduled batch jobs.
- Emergency Console Access: Check the setting allowing administrators in the Claude Console or owners in claude.ai to reach settings from unlisted networks to maintain an administrative rescue path.
- Static Proxy Accounting: Services utilizing bridge.claudeusercontent.com, Cowork, or Office add-ins that route via alternate gateways must have those egress IPs included.
- Active Background Tasks: If a user is restricted while a conversation task is executing, the process can persist for up to 24 hours, though the user cannot control it from an unapproved IP.
- Support Remediation: If team administrators lose access entirely, resolving the lockout requires contacting Anthropic support.
To maintain audit transparency when keeping the administrator rescue path open, teams can monitor updates through the Compliance API Activity Feed. Every modification to approved IP lists generates an auditable record.
Admin API Remote Session Revocation
Complementing IP restrictions, the 2026-11-05 update introduced an Admin API beta endpoint for Claude Enterprise organizations to force immediate session termination across all devices. When an enterprise account suffers credential compromise or an employee leaves an active device unmonitored, administrators can programmatically terminate active tokens without removing the user from the organization directory.
curl -X POST https://api.anthropic.com/v1/organizations/users/user_01AbCdEfGhIjKlMnOpQrSt/sign_out \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2026-10-21"
Invoking this sign-out endpoint invalidates web, desktop, and mobile tokens immediately. Because the user record remains active, valid corporate credentials allow the member to sign back in later through standard identity channels. Calling this endpoint requires an Admin API key carrying the write:members scope. For cost tracking associated with user activity, administrators can connect the Anthropic Usage and Cost Analytics API.
Deployment Checklist for Engineering Teams
Before enforcing allowlists on production Claude Platform accounts, engineering teams should conduct a discovery phase to identify all external dependencies. Missing a single scheduled pipeline will cause failure once the 15-minute propagation completes.
- Audit all CI runners, staging environments, production servers, and proxy egress endpoints across both IPv4 and IPv6.
- Verify that external key management configurations—such as AWS KMS, Cloud KMS, or Azure Key Vault—are properly scoped to your infrastructure.
- Ensure the bypass checkbox allowing administrators to reach settings pages from any address remains enabled during rollout.
- Monitor server logs for HTTP 403 permission rejections across the Messages and Message Batches endpoints.
- Combine network rules with ephemeral keys or Workload Identity Federation rather than relying solely on static tokens.
Review token consumption and budget guardrails via the LLM API cost calculator to monitor how background tasks and automated deployments perform within secured perimeters.