Skip to content
NewsProduct launch

Anthropic adds a Claude Enterprise security center and moves inference hooks to general availability

· by

The short version

Anthropic shipped a security center, generally available inference hooks, and centralized admin controls for Claude Enterprise on November 5, 2026, then added beta cost-code spend tagging the next day.

Anthropic adds a Claude Enterprise security center and moves inference hooks to general availability

Anthropic took inference hooks, the prompt-screening feature it launched in beta three months ago, out of beta on November 5, 2026, and paired that move with a new audit dashboard and a centralized admin console, per Anthropic's announcement.

What

The security center is a new page under Organization settings that checks a company's Claude configuration against Anthropic's recommended settings every time the page loads, flagging what differs and linking straight to the setting that needs changing, per Anthropic. Admins can download the results as a report for auditors. Inference hooks, which route prompts and tool calls through a company's own data-loss-prevention server before Claude acts on them, now ship in two generally available forms: prompt hooks and pre-tool-call hooks, both previously beta since Anthropic's August 2026 launch. Moving them to general availability removes the single biggest caveat security teams raised about that August launch. Centralized management gives a parent organization its own console at admin.claude.com for single sign-on, SCIM directory sync, and Compliance API keys across every Claude Enterprise and Claude Platform org underneath it, replacing per-org configuration. New access controls let admins force a single sign-out across a member's sessions through the Admin API, manage their own IP allowlist without going through Anthropic, and block personal Claude accounts on managed desktops.

A second, narrower feature followed one day later. On November 6, Anthropic enabled project tags in beta, letting eligible Claude Enterprise organizations tag a project with an existing cost code, such as a work-breakdown-structure number or client-matter code, then view spend grouped by that tag under Organization settings, the company says. A project can carry up to 16 tags, and an owner sets up the tag names while anyone with edit access to a project can assign the value.

Project tags skip the API, Bedrock, and Vertex entirely

The spend-tagging feature has a real gap operators should check before they rely on it for chargeback. According to the same support article, project tags only track usage inside a Claude project, and several major surfaces are excluded outright: the Claude API, the Claude Console, Claude Code running through Amazon Bedrock or Google Cloud Vertex AI, Claude for Office, Claude in Chrome, and Claude in Slack and Microsoft Teams. Claude Code in a terminal only gets tagged after a developer manually sets an attributionProject value in a settings file, and that step has no warning if skipped, so usage silently falls outside the report. A finance team relying on Claude Code's cost-code numbers for chargeback should confirm with IT that attributionProject is actually set in every repository's settings file before trusting the dashboard. An organization that runs most of its Claude Code traffic through a cloud-provider gateway rather than Anthropic's own Enterprise plan will see little of that spend show up by tag at all. The feature also is not available to organizations running HIPAA configurations, customer-managed encryption keys, or zero data retention.

Sources