Research
Anthropic Ships Inference Hooks: Enterprise DLP Servers Now Vote Allow/Deny on Every Claude Prompt Before Inference
On August 5, 2026 Anthropic launched inference hooks in beta for Claude Enterprise, routing each prompt and tool-call response over a signed WebSocket to the customer's own security server for an allow-or-deny verdict before the model generates anything. Coverage spans Claude chat, Claude Code, Claude Cowork, and tool calls made through MCP connectors, skills, and plugins, with named integrations for Netskope, Palo Alto Networks, Proofpoint, and Zscaler, plus shadow mode, role-based exclusions, and percentage rollouts. The current limitation matters for builders: verdicts are binary — the server can block a prompt but cannot redact or rewrite it — so any redaction logic still has to live client-side.
↳ Follow the thread