The biggest Claude Enterprise security update so far landed on November 5, and the part most teams will care about is that prompt and tool-call inspection is now generally available. Anthropic also added a settings checker, a company-wide admin console, and endpoint monitoring for Cowork. Several pieces are still beta, and a few have gaps you should know about before you rely on them.
Not an admin? Skip to the FAQ. The short version is that this changes what your IT team can see and stop, not what Claude can do for you.
What Anthropic shipped for Claude Enterprise security
Here are the quick facts, straight from Anthropic’s announcement:
- Announced on November 5, 2026 for Claude Enterprise customers.
- Inference hooks are the only generally available piece. Everything else is beta.
- Cowork endpoint monitoring also covers Claude Team plans.
- No price change was announced.
The table below shows what’s live and what’s still in testing.
| Feature | Status | What it does |
|---|---|---|
| Inference hooks | Generally available | Send prompts and tool calls to your existing security tool before they run |
| Security center | Beta | Compares your settings with Anthropic’s recommended setup and exports a report |
| Parent console | Beta | One admin console for SSO, directory sync and domains across organizations |
| Member sign-out (Admin API) | Beta | Ends a person’s sessions on every browser and device |
| Self-serve IP allowlist | Beta | Admins manage approved IP addresses without contacting Anthropic |
| EDR in Cowork | Beta | Runs your endpoint security sensor inside the Cowork virtual machine |
Most of this is plumbing for security teams. But it matters because AI tools now touch files, bank data and code, and companies want the same oversight they already have for email and laptops.
Inference hooks: your DLP tool gets a say before Claude answers
Inference hooks connect Claude to the data loss prevention (DLP) tools a company already pays for. A prompt hook sends each prompt to your security tool for a check. A pre-tool-call hook does the same before Claude runs a tool.
The appeal is simple. Your existing rules, like blocking customer card numbers from leaving the building, carry over without a rewrite.
According to an independent analysis from Beri, there are catches. The default setting lets requests through if your DLP server is slow or down, with a 5-second timeout. Images aren’t inspected, voice is only partly covered, and hooks can allow or deny but can’t redact.
If you turn hooks on, change the failure setting to block before you call it done. Otherwise an outage quietly turns your protection off.
Security center and parent console: one place to see the whole picture
The security center is a single settings page that compares your chat, Cowork and Claude Code configuration with Anthropic’s recommended setup. For each control it shows the current value, the recommended value and a link to change it.
Admins can also record accepted risks, and the downloadable report separates automatic checks from items an owner confirmed by hand. Anthropic says plainly that it is guidance, not a compliance certification.
The parent console is for bigger companies with several Claude organizations. It manages single sign-on, directory sync, domain verification and Compliance API keys in one place at admin.claude.com.
One warning: parent setup cannot be undone. Pick your parent admins before you start, and treat parent-level Compliance API keys as highly sensitive, since they can read data across every organization.
Sign-out, IP allowlists and Cowork monitoring: handy, with limits
The new Admin API call signs a person out of Claude on every browser and device without removing them from the organization. It’s useful in the first hour of an incident, when a laptop goes missing or an account looks hijacked.
It isn’t a full offboarding, though. API keys, cloud sessions already running and scheduled routines keep going, so you still need to revoke those separately.
The self-serve IP allowlist lets admins manage approved addresses on their own. Claude Desktop settings management also lets owners block personal accounts on managed computers.
The Cowork piece lets teams run their own endpoint detection and response (EDR) sensor inside the virtual machine Cowork uses on a person’s computer. It only watches and reports. It doesn’t block anything, and it doesn’t cover cloud Cowork sessions.
The data that sensor sees can include prompts and files inside the VM. Check with your EDR vendor on where that data goes before rollout.
What this means for everyday Claude users
If you use a work Claude account, expect more visibility on your side of the screen. Your company may scan prompts, restrict outside accounts in Claude Desktop, or limit which folders Claude can use.
Personal accounts aren’t affected. If you’re curious how Anthropic handles sensitive data on the consumer side, our guide to Claude Personal Finance covers what the new bank-connection beta can and can’t see.
Security teams worried about secrets leaking into AI tools may also want our checklist after the Northflank breach, which explains why rotating AI keys comes first.
Frequently Asked Questions
Is Claude Enterprise security available on Pro or Max plans?
No. These controls are for Claude Enterprise, except Cowork endpoint monitoring, which also covers Team plans.
Do inference hooks read my private chats?
Hooks send prompts to a security tool your employer chooses. So yes, on a work account your company can inspect them. Personal accounts don’t use hooks.
Does the security center make a company compliant?
No. Anthropic calls it guidance, not a certification. It shows how your settings compare with its recommendations.
Can an admin lock someone out instantly?
Partly. The Admin API ends browser and device sessions. API keys, running cloud sessions and scheduled routines keep working until someone shuts them down.
Does the EDR sensor block threats in Cowork?
No. It runs in detection mode and reports activity to your security console. Blocking needs a separate process.
My take: the hooks are the real news, and they’re worth turning on this week. Just flip the failure setting to block and test it with your DLP server switched off. Treat the beta pieces as a good checklist for now rather than proof you’re covered, and keep your usual offboarding steps in place.
Leave a Reply