On November 5 Anthropic announced a package for security teams, and it is larger than the usual enterprise footnote: a security center, general availability for inference hooks, centralized management across organizations, and three new access controls. Several of them are in beta.
The security center compares your settings with Anthropic’s recommendation
There is a new page in Organization settings that shows the security settings for chat, Cowork and Claude Code together, instead of setting by setting. For each control you get the current value, the recommended value where there is one, and a link to where you change it. Anything new or recently changed is flagged. The check runs on every page load, and organizations with the Compliance API can read the same results through it.
Alongside that come attestations: a shared workspace where admins walk through a security review once together, record the state of each control, accept risks and flag issues. The page downloads as a report, and the report separates what was checked automatically from what an Owner confirmed.
Inference hooks are now generally available
Anthropic introduced inference hooks in August; they hand data to the security tools an organization already runs. Both kinds leave beta now: prompt hooks send prompts from supported Claude products out for a check, pre-tool-call hooks check tool calls before they run. The policies stay in your own DLP tool.
One console across several organizations
Centralized management gives the parent organization — the one that represents a company across its Claude Enterprise and Claude Platform organizations — its own console at admin.claude.com. Single sign-on, directory sync via SCIM, domain verification, groups and Compliance API keys live there, set once for every organization underneath.
Two roles come with it: parent organization owners can do everything, including role assignment; parent organization admins can do everything except that. Several identity providers are allowed, each with its own SSO and SCIM configuration, which is aimed at acquired companies that want to keep their provider or move one domain at a time. Members, roles and the organization-level controls stay with each organization.
The three access controls are already in the release notes
Two of them have been in the developer changelog since November 5. The IP allowlist can now be managed in the Claude Console yourself, in beta, with IPv4 and IPv6 addresses or CIDR ranges, up to 1,000 per organization; requests from addresses that are not on the list are refused, which limits what a leaked API key is worth.
The second is an endpoint: POST /v1/organizations/users/{user_id}/sign_out, in beta, and every request needs the admin-user-sign-out-2026-10-08 beta header. The call ends sign-in sessions and trusted devices in every browser and on every device, plus the member’s OAuth tokens for the organization. The member stays in the organization and can sign in again right away. The same capability showed up on November 5 in the Python SDK 2.6.0 and the TypeScript SDK 1.6.0.
The third control is aimed at shadow IT: on managed computers, personal and other outside Claude accounts can be blocked in Claude Desktop.
A page of recommendations is not a certificate
Anthropic says it in the post: the security center is guidance based on its own recommendations, not a certification and not a guarantee that any law or standard is met. Hand that report to an auditor and you are handing over a self-assessment with two kinds of checkmarks in it. That is more than existed before, and less than the page looks like at first glance.