AI news for business · checked November 12, 2026

Anthropic adds a security center, generally available inference hooks, centralized management and new access controls to Claude Enterprise. Several parts are still in beta.

On November 5, 2026, Anthropic introduced new security and admin features for Claude Enterprise. At the center is a security center that compares an organization's settings for chat, Cowork and Claude Code with Anthropic's recommended configuration. It is joined by inference hooks for existing security tools, centralized management across several organizations and additional access controls. The vendor facts come from the announcement; the working steps below are our own assessment.

LOL.MARKETING / CONNECTING INTELLIGENCE

The essential answer

The security center (beta) shows the current value and, where one exists, the recommended value for each control, with a link to the setting, and can be downloaded as a report. It is explicitly not a certification or proof of compliance. Prompt hooks and pre-tool-call hooks are generally available. Centralized management, member sign-out via the Admin API, the self-serve IP allowlist and desktop settings are in beta.

01 / FIT

A good fit when

  • Security, IT and data protection teams at companies on Claude Enterprise, especially with several Claude organizations or managed computers.

02 / LIMITS

Not the first choice when

  • Claude Pro, Max or Free users: the announcement targets Claude Enterprise; only the EDR integration also lists Claude Team.

What is the Claude Enterprise security center?

The security center is a new page in Organization settings. Security teams see the security settings for chat, Cowork and Claude Code on one page instead of opening each setting individually. For every control, the page shows the current value, the recommended value where Anthropic defines one, and a link to the place where the setting can be changed. New or recently changed settings are flagged.

The check runs again every time the page loads, so automatically checked controls reflect the current state. Organizations with the Compliance API can read the same results through it. With attestations, admins work through a security review together: they record the state of each control, accept risks and flag open issues in one place.

Does the report replace proof of compliance?

No. The page can be downloaded as a report and shared with auditors or other stakeholders. The report marks which controls were checked automatically and which an Owner confirmed. Anthropic, however, explicitly describes the security center as guidance based on its own recommendations, not as a certification and not as a guarantee of compliance with any law or standard.

For companies in Germany and the EU, this means the report can supplement data protection or ISMS documentation, but it replaces neither your own risk assessment nor your records of processing or coordination with your data protection officer. Decide who assesses deviating values, and document why you deliberately do not adopt a recommendation. That keeps the decision traceable even if recommendations change later.

How do inference hooks connect your own DLP tools?

Inference hooks connect Claude to the data security tools an organization already uses, so data loss prevention policies also apply to prompts and tool calls. There are two kinds, both now generally available: prompt hooks send prompts from supported Claude products to the security tool for a check. Pre-tool-call hooks check tool calls in supported Claude products before they run.

The hooks work with security vendors; existing policies stay in place and are applied to Claude. In practice, clarify beforehand which of your products count as supported, which data the security tool receives in the process and who handles false positives. A test with non-sensitive sample data shows whether blocks and alerts arrive the way your security team expects.

What does centralized management at admin.claude.com offer?

Centralized management (beta) gives a parent organization, which represents the company across its Claude Enterprise and Claude Platform organizations, its own console at admin.claude.com. Single sign-on, SCIM, domain verification, groups and Compliance API keys are managed there once for all organizations under the parent. There are two roles, parent organization owner and parent organization admin; only owners manage role assignments. Several identity providers with their own SSO and SCIM configuration are possible.

Two points deserve attention. First, according to Anthropic, setup cannot be undone; the parent admins should be settled beforehand. Second, a parent role does not grant access inside an organization by default, but Compliance API keys created at the parent level can read data across organizations. Issue such keys sparingly, document their purpose and involve data protection.

Which new access controls are available?

Security teams can sign a member out with a single Admin API call without removing them from the organization. This ends sign-in sessions on every browser and device; apps such as Claude Code have to sign in again. API keys, cloud sessions already running and scheduled routines do not end, and the member can sign in again afterwards. A real offboarding or incident process therefore needs additional steps.

Enterprise admins can also manage their IP allowlist themselves without going through Anthropic. Admins can set the most-used Claude Desktop settings without a device management push and block personal and other outside Claude accounts in Claude Desktop on managed computers to help prevent shadow IT. Owners can also limit which workspace folders people can use with Claude. Member sign-out, the IP allowlist and desktop settings are in beta.

What does the EDR sensor in the Cowork VM do?

When Cowork and unified Claude work locally, shell commands and code run in an isolated virtual machine on the computer. Admins can now run their organization's EDR sensor inside this VM; it reports activity to the security console already in use. This covers the local VM, not Cowork sessions in the cloud. Before each installation, Claude verifies the signature of the sensor package.

The sensors run in detection mode: they report activity but do not block it. Setup runs through device management on macOS and Windows and is in beta for Claude Team and Enterprise, and for the Claude desktop app on Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. Anthropic lists the supported EDR vendors in its Help Center.

How should companies proceed now?

According to Anthropic, the features are available for Claude Enterprise, with some beta exceptions listed in the Help Center; the announcement does not list regional restrictions. We recommend a short, structured review before changing settings.

  • Open the security center, record deviations from the recommendation and name an owner for each item.
  • Use the report as a working basis for audits and data protection, never present it as a certificate.
  • Decide on parent admins before centralized management, because setup cannot be undone.
  • Extend the offboarding process: sign-out does not end API keys, running cloud sessions or routines.
  • Trial inference hooks with non-sensitive test data first and record rules on personal data in the AI policy.

Keep it verifiable

Primary sources

The next sensible step

Has your Claude configuration been reviewed?

Eight steps from a general interest in AI to a clearer decision for your business.

Start AI System Check
FreeProvider-neutralNo credentials