Wain AI/Tech Blog

AI news and trends worldwide, updated nearly every day

Claude Enterprise Adds a Security Center That Compares Settings With a Recommended Configuration, Plus Parent-Organization Centralized Management in Beta - Enrolling in Centralized Management Is Irreversible

Claude Enterprise Adds a Security Center That Compares Settings With a Recommended Configuration, Plus Parent-Organization Centralized Management in Beta - Enrolling in Centralized Management Is Irreversible

On November 5, 2026, Anthropic added a security center and centralized management (admin.claude.com), which handles SSO and SCIM at the parent-organization level, to Claude Enterprise in beta. A setting to block personal accounts in Claude Desktop was also added, and inference hooks were described as generally available. Enrollment in centralized management cannot be undone, and a parent organization with a HIPAA-enabled child organization cannot enroll.

On November 5, 2026, Anthropic added a batch of security and administration features to Claude Enterprise1. There is a security center that sets an organization’s settings against Anthropic’s recommended configuration, centralized management, which gives a parent organization spanning several Claude organizations a separate administration console, and new access controls, including a setting that blocks personal accounts in Claude Desktop. Inference hooks, released in beta in August, were described in this announcement as generally available.

Availability differs by feature. The security center and centralized management are in beta, and so are Admin API sign-outs, IP allowlists that admins configure themselves, and control of Claude Desktop through Organization settings. Inference hooks alone have reached general availability1.

What follows goes through each feature, focusing on the conditions worth checking before turning it on or enrolling.

The security center only shows; it does not change settings

The security center lives on one page inside Organization settings. It displays the organization’s settings across chat, Claude Cowork, and Claude Code next to Anthropic’s recommended configuration2. It opens from the account menu under “Organization settings” and then “Security posture.”

Controls are split into four groups: Org, chats, Cowork, and Claude Code. For each one, the page lists what the setting is now, what Anthropic recommends (if it has a recommendation), a pass or fail result, and a pointer to the page where it is changed. Each control has one of four statuses:

  • Passed: in line with the recommendation
  • Failed: differs from it; the page explains why that matters and which page to visit
  • Not attested: a person needs to confirm it
  • Shown for information: displayed without a judgment

According to Anthropic, the check reruns whenever the page is opened, so results for automatically evaluated items match the live configuration1.

Some settings, however, cannot be read automatically. Anything controlled somewhere other than Claude’s own admin pages, such as device management (MDM), Claude Code managed settings, or a network proxy, appears as “Check manually,” and someone fills it in through “attestation,” recording its current state2. The page also records who attested and when. Organizations that enforce many controls on the device side will likely end up relying on human records rather than automatic checks for more of their items.

There are three operational limits2:

  • View-only: settings cannot be changed from this page; each one is changed on its own page
  • Owners and the Primary Owner only: admins and regular users cannot open it
  • Not a certification: a PDF report can be downloaded for auditors, and it distinguishes items the page verified on its own from items an Owner vouched for manually, while stating that it is not a security certification

Anthropic also calls it guidance drawn from its own recommendations and says it does not promise that an organization meets legal or regulatory requirements1. Organizations that have the Compliance API enabled can pull these results into their own tools through that API.

Centralized management moves SSO and domains to the parent

Centralized management gives the “parent organization,” which represents the whole company, a dedicated admin console at admin.claude.com3. Under the parent sit “child organizations,” such as Claude Enterprise organizations, Team plans, and Claude Platform organizations bought directly from Anthropic.

Enrolling does not merge the children. Every child keeps its data boundary, along with its users, roles, and projects3. What changes is that company-wide settings, previously operated from each child’s admin settings, are handled once in the parent console.

The main settings transferred to the parent are3:

  • SSO and Require SSO, directory sync (SCIM), and domain verification
  • Connector domain restriction and restricting organization creation
  • Compliance API keys and groups

Security controls set per organization, for example the IP allowlist and how long sessions last, remain in the hands of each child’s admins.

Among the new controls, a single parent can have several identity providers attached. Anthropic says this lets an acquired company keep the provider it already uses, or lets a company switch providers gradually, domain by domain1.

Two parent roles are added. A parent owner controls the whole parent console, including who receives the owner and admin roles, while a parent admin has the same reach minus those role decisions1. Holding a parent role gives no default entry into any child organization. Keys for the Compliance API issued at the parent level are different: they can read data from all of the children. Deciding who gets parent roles therefore also means deciding who can issue those keys.

Conditions to check before enrolling

The Help Center warns that enrollment is permanent and that it may be finished the moment setup begins3. Setup can be started at admin.claude.com by an owner of any child. If the person starting it administers all of the children, enrollment takes effect immediately. If not, each child signs off on the proposed parent admins, and enrollment finishes once the final child approves. Editing the list of proposed admins resets the approvals for every child, and if a week passes after the most recent edit without a child’s approval, an owner from a different child can sign off for it.

During the beta, these conditions apply3:

  • Parents that cannot enroll: any parent that has a child organization with HIPAA enabled, and any parent whose only children are Team plans
  • Children that cannot join: Claude Platform organizations bought on the AWS or Azure marketplaces
  • Creating new child organizations: goes through the account team; there is no self-serve option
  • Billing: invoicing does not change on enrollment
  • Group visibility: there is not yet a control for limiting a group to particular children, so for now all groups under the parent are visible to the admins of every Enterprise child

Networks that use tenant restrictions must put the parent organization’s ID on the allowed-ID list sent in the anthropic-allowed-org-ids header, and do so ahead of setup3. Otherwise, people holding parent roles cannot manage the parent console from that network.

The Claude Desktop personal-account block lapses after 30 days

On access controls, admins can now change the main Claude Desktop settings from Organization settings without pushing them through MDM1. One of them shuts out personal accounts, along with any organizations not on the allowed list, on managed computers. Anthropic positions this as a way to prevent shadow IT.

The Help Center describes how it works and where it stops4:

  • When it starts: once a member has used Claude Desktop on a machine to sign in to the organization’s account, the block takes effect the next time the app is opened. To have it in place even before the first sign-in, configure forceLoginOrgUUID through MDM
  • Not permanent: it lifts if nobody uses the organization on that machine for 30 days, and whoever is using the machine is able to clear it
  • Device policy wins: when a local policy on the machine sets the same key, it takes precedence even if it is looser
  • Version: requires Claude Desktop 4.29.0 or later

Being able to do this from Organization settings alone is convenient, but the Help Center’s description suggests MDM is still needed on computers where the block has to hold.

The same settings page lets owners restrict the workspace folders available to Claude. A note explains, though, that marking a folder “Read only” covers only Claude’s file tools; hooks, MCP servers, and commands launched from the Code tab are still able to modify that folder and get to others4.

Signing a member out does not stop API keys or running work

One call to the Admin API logs a member out across all their browsers and devices while leaving their membership intact1. Apps such as Claude Code then require a fresh sign-in. Existing API keys, cloud sessions in progress, and scheduled routines keep going, and nothing prevents the member from logging back in afterward. Stopping API keys or cloud sessions as well would appear to take separate steps.

For IP allowlists, Enterprise admins no longer need to go through Anthropic to configure and maintain one1. On the same day, November 5, the Claude Platform release notes announced that an organization’s IP allowlist can now be set in the Claude Console as well, in beta7.

Inference hooks move from beta to general availability

Inference hooks were released in beta in August. They send prompts to the organization’s own security server and wait for it to return allow or deny before proceeding. For tool calls, the August announcement said tool output would be inspected before being handed back to the model6.

This announcement lists two kinds of hooks and says both are generally available1: prompt hooks, which send prompts to a security tool for a check, and pre-tool-call hooks, which check tool calls before they run. The Help Center says they cover all Claude Enterprise products, including Claude, Claude Code, and Cowork, and that Owners and Primary Owners turn them on and manage them5. Since the check runs in Claude’s own infrastructure and not on employees’ machines, nothing has to be installed on them.

Before rushing to enroll

These features follow October’s shared group budgets and usage-analysis reports as management tools for organizations running Claude Enterprise company-wide. After cost management, the changes now reach configuration review, centralized identity, and limits on use from company computers.

Centralized management is where the decision is least obvious. For a company with several child organizations, it puts company-wide settings in one place, but enrollment is permanent, and a company with a HIPAA-configured child cannot enroll at all. Who will hold the parent owner and admin roles, whether the tenant-restriction header has been updated first, and whether it is acceptable for every child’s admins to see every group: checking these three points before starting setup seems the safer approach.

Because the security center is view-only and changes nothing, it is easy to try simply to see where things stand. Keep in mind, though, that automatic checks are limited to settings inside Claude’s admin settings, and controls on the device or network side are left to human attestation.

Sources

  1. More visibility and control over Claude Enterprise for security teams - Anthropic (November 5, 2026)
  2. Review your organization’s security posture with the security center - Claude Help Center (checked November 7, 2026)
  3. Get started with centralized management - Claude Help Center (checked November 7, 2026)
  4. Enterprise configuration for Claude Desktop - Claude Help Center (checked November 7, 2026)
  5. Inference hooks overview - Claude Help Center (checked November 7, 2026)
  6. Inference hooks: inline data loss prevention for Claude Enterprise - Anthropic blog (August 5, 2026)
  7. Claude Platform release notes - Anthropic (entry for November 5, 2026)

We publish the latest AI news nearly every day.

Subscribe via RSS Get new posts the moment they go live.

Search other keywords →