Claude Enterprise adds security controls, with different limits for hooks, sign-out and monitoring

Anthropic announced a set of Claude Enterprise security and administration updates on November 5, including a security center, centralized management and generally available inference hooks. The release gives security teams more ways to inspect configuration and intervene in agent activity, but the features have different availability stages and enforcement powers. Anthropic announcement
That distinction matters during deployment and incident response. A dashboard, a policy decision and an endpoint sensor answer different questions. None should be assumed to replace the others.
Hooks can check actions before they run
Anthropic’s current help documentation says inference hooks send prompts and tool calls to an organization-controlled server for an allow-or-deny decision. Claude waits for that answer before continuing. The checks operate within Claude’s infrastructure and can cover tools exposed through skills, plugins and connectors. Owners and Primary Owners manage the feature. Inference hooks overview
For administrators, the important implementation work is deciding what the server should permit, what it should reject and how exceptions will be reviewed. A policy integration can only enforce the rules it actually receives. Test routine work as well as prohibited cases, so a rollout does not quietly turn into either a blanket block or a permissive default.
Configuration guidance is not certification
The new security center compares settings with Anthropic’s recommended configuration and distinguishes automatic checks from administrator attestations. It is in beta, and the company explicitly describes its findings as guidance rather than a compliance certification. Security center announcement
Centralized management is also in beta. It provides a parent-level console for identity settings while preserving separate child organizations. A parent administrator does not automatically gain access inside each child. However, parent-level Compliance API keys can read across organizations, a separate permission boundary that deserves review. Centralized management documentation, Release details
Teams should map who controls identity, who can read compliance data and who owns each individual workspace before enrollment. Consolidating administration is useful only if those responsibilities remain understandable.
Signing out is narrower than stopping everything
The new Admin API sign-out action ends a member’s login sessions across devices. Anthropic says it does not revoke API keys, stop existing cloud sessions or cancel scheduled routines. The member can also sign in again afterward. Sign-out scope
An incident checklist should therefore treat session termination, credential revocation and stopping ongoing work as separate decisions. Assuming one sign-out call performs all three could leave the response incomplete.
Endpoint visibility has its own boundary
The EDR integration runs a supported vendor’s sensor inside Cowork’s local virtual machine. It is a beta for Team and Enterprise on supported desktop platforms and does not cover Cowork sessions running in Anthropic’s cloud. Sensors report activity rather than block it. EDR documentation, Detection-mode limitation
The practical gain is a more complete set of controls around agents. A sound rollout still needs an explicit coverage map: which products are checked, which actions are blocked, which events are only reported and who responds when something goes wrong.
Sources and citations
Revision history
Published .