Anthropic Ships Claude Security Center, Inference Hooks GA
TL;DR
- Inference hooks run server-side on Anthropic's infrastructure, so one policy covers Claude.ai, Claude Code, and Claude Cowork without endpoint software on user devices.
- Netskope, Palo Alto Networks, Proofpoint, and Zscaler ship pre-built integrations, letting enterprises extend existing email and web DLP policies to Claude without custom webhook code.
- Pre-tool-call hooks intercept MCP, skills, and plugin responses before they reach the model, closing an inspection gap that prompt-only scanning misses.
Anthropic's Claude enterprise security update, posted November 5, 2026, introduces an in-product Security Center that lines up a customer's Claude settings against the company's own recommended configuration and flags the gaps.
The page covers chat, Cowork, and Claude Code together. In the company's words, it lets admins "review Claude's security settings for chat, Cowork, and Claude Code on one page, instead of setting by setting." It is in beta and can be exported as an audit report; organizations on the Compliance API get the same results through that interface.
Inference hooks are now generally available. Prompt hooks ship outgoing prompts to a customer's existing security tool for inspection; pre-tool-call hooks intercept tool calls before they run. The pitch, in the post's words, is that "data loss prevention (DLP) and security policies can apply to prompts and tool calls as people work" without the team rewriting them. A separate console at admin.claude.com handles parent-organization management across Claude Enterprise and Claude Platform orgs, including single sign-on, SCIM directory sync, domain verification, and Compliance API keys.
EDR integration inside Cowork is in beta on macOS and Windows. Admins can run their organization's sensor inside the Cowork virtual machine on user devices, and Claude verifies the sensor package's signature from the EDR vendor before installing it. The behavior has a limit stated plainly: "Sensors run in detection mode: they report activity to your security team and don't block it." Claude Desktop can also be locked down on managed computers to block personal and other outside Claude accounts, aimed at shadow IT.
The release lands a day after a Claude Code bug report alleged that user emails were leaking through a User-Agent header, one of 368 Anthropic stories on our tracker over the last 90 days.
What others are reporting
-
The Next Web Read →
Frames inference hooks as Anthropic's direct answer to AI security startups, names Netskope/Palo Alto/Zscaler integrations, and connects the launch to White House frontier-model policies.
The same check runs on tool calls: when Claude calls a tool through MCP, skills, or plugins, the tool's response is checked before it reaches the model.
-
Unite.AI Read →
Covers webhook protocol design and enforcement placement on Anthropic's servers; maps beta vs. GA scope, noting API-direct and Bedrock/Google Cloud deployments remain excluded.
Inference hooks moves the checkpoint onto Anthropic's servers, after a request leaves the client and before inference runs
-
Superpower Daily Read →
Explicitly distinguishes GA features from beta, and notes local monitoring is detection-only and excludes cloud sessions, clarifying what this release does not yet cover.
Prompt hooks send prompts to the organization's security tool. Pre-tool-call hooks check proposed tool calls before they run. Both are generally available.
-
Blockchain.News Read →
Highlights the EDR integration path for local VM monitoring and the signed HTTPS webhook delivery protocol as technical differentiators versus network-layer AI firewalls.
Inference hooks, now generally available, connect Claude to existing data security tools, enabling organizations to enforce data loss prevention (DLP) and other security policies.
Originally reported by claude.com
Read the original article →Original headline: Anthropic Ships Claude Enterprise Security Center, Makes Inference Hooks GA for Prompt and Tool-Call Scanning