Anthropic has given Claude Enterprise its most complete security package yet, announcing new controls on November 5 that let security teams see how the assistant is configured, apply their own data-protection rules to its work, and manage access across a whole company from one place. The headline change is inference hooks, now generally available, which send prompts and tool calls to a customer's own security tools for a check before Claude answers or acts. A security center, a central admin console, and new access controls arrive in beta alongside them.
A security center that grades the setup
The security center is a new page in Organization settings that lines up an organization's Claude settings for chat, Cowork, and Claude Code against Anthropic's recommended configuration. For each control it shows the current value, the recommended value where one exists, and a link to the setting that changes it.
The page also flags settings that are new or recently changed, and teams can download it as a report to share with auditors. Anthropic is explicit about the limit: the center is guidance based on its recommendations, not a certification or a guarantee of compliance with any law or standard.
Inference hooks: the check before Claude acts
Inference hooks are the one part of the package that is finished: both kinds are now generally available on Claude Enterprise. Prompt hooks send a user's prompt to the organization's security tool for a check before Claude generates a reply, and pre-tool-call hooks inspect a proposed tool call before it runs.
The point is that a company keeps the data-protection policies it has already written and applies them to Claude, instead of rebuilding them inside the assistant. Hooks work with the security vendors a team already uses, and the check runs inside Claude's infrastructure rather than on each employee's device.
One console, and new ways to lock the doors
Centralized management gives a parent organization its own admin console at admin.claude.com, for companies running several Claude Enterprise and Claude Platform organizations. Admins manage single sign-on, directory sync, domain verification, groups, and Compliance API keys once, instead of repeating the setup in each organization.
The console adds two parent roles, and a company can connect more than one identity provider, so an acquired business can keep its own sign-in while the parent still sees every organization in one place. Setup cannot be undone, so Anthropic tells admins to choose their parent administrators before they start.
The access controls cover the everyday risks. Security teams can sign a member out of Claude with a single Admin API call, ending sessions on every browser and device, though it does not end API keys or cloud sessions already running.
Enterprise admins can now manage their own IP allowlist, set the most-used Claude Desktop settings without a device-management push, and block personal and other outside accounts on managed computers to stop shadow IT. Owners can also limit which workspace folders people may use with Claude.
The Endpoint Detection and Response integration puts a company's own sensor inside the virtual machine where Cowork and the unified Claude app run shell commands and code on the user's computer, closing a visibility gap security teams had asked Anthropic to fix. The sensor reports activity to the console the team already monitors.
The sensor runs in detection mode only, reporting activity without blocking it, and Claude verifies the sensor package's signature from the vendor each time the virtual machine starts. The integration covers the machine on the user's computer, not Cowork sessions in the cloud, and it is in beta on macOS and Windows.
Our Read
The package reads as Anthropic answering the objection that has slowed Claude in companies: not the model's ability, but whether security teams can see and control it. Inference hooks are the load-bearing piece because they are the only part that is generally available, and they reuse tools a company pays for instead of asking it to buy a new one.
The rest is beta, so the real test is whether the security center and the central console survive contact with messy enterprise setups. The Claude Code email leak showed what happens when a control is missing, and the hooks are Anthropic's strongest answer so far.
Why It Matters
Enterprise buyers decide on controls as much as on capability, and this package removes the most common reasons to say no to Claude at work. A security team can now check prompts and tool calls against its own data-loss rules, see the whole configuration in one place, and manage access company-wide.
For Anthropic, that is the point: the enterprise tier only grows if the people who guard company data are comfortable letting Claude touch it. The timing fits a company heading for the public markets, where enterprise revenue is the story investors will judge it on.
Background
Anthropic has spent 2026 pushing Claude from a chat assistant into workplace infrastructure, with Cowork, Claude Code, and an enterprise tier aimed at companies. The same week brought a consumer launch, Personal Finance in Claude, showing the two tracks running side by side. The security package is the administrative layer that push was missing.
0 Comments