Browse Docs
ONLINE DOCUMENTATION

Enforcing AI Protections for Claude

Overview

DefensX AI Protections help organizations prevent sensitive data from being inadvertently shared with generative AI platforms like Claude. The concern here is straightforward. Claude includes a "Help improve Claude" option in settings, which, when left enabled, allows Anthropic to use conversation data for model training purposes. For businesses handling confidential information, this represents a real data exposure risk that needs to be managed proactively.

How it works?

Unlike some services that provide network-level controls, Claude has no native mechanism such as a DNS signal or HTTP header that administrators can use to enforce data-sharing preferences across their user base. To work around this, DefensX takes a browser-level approach through its extension. During each Claude session, the extension monitors the page and checks whether the "Help improve Claude" setting is active. If it is, all requests to Claude are blocked and the user is presented with a clear message instructing them to turn the setting off before they can proceed.

Note
 Disabling this setting is an action only the user can take. DefensX handles the detection and enforcement side, guiding users through the process with an on-screen prompt.

Configuration

Tip
 DefensX AI Protections feature is available for CORE+ and higher packages.

To activate this protection, go to Settings → AI Protections, locate the Claude, and switch on the "Disable Data Sharing for Model Improvements" toggle. Allow up to 5 minutes for the change to propagate due to caching.

Screenshot%202026 04 24%20at%2013.57.41

Blocking Fat Clients

Browser extension controls only go so far; they won’t cover users accessing Claude through browsers or applications that don’t have the DefensX extension installed. To close this gap, DefensX Web Filter policies can be used to enforce DNS-level restrictions.

  • Restrictive policy: Set up a web filter policy targeting Agent DNS and Cloud DNS (excluding browsers) that blocks either the Artificial Intelligence category or a custom URL group built around *.claude.ai. Position this policy, so it runs ahead of any more permissive rules. This way, Claude is blocked at the DNS level for any device that isn’t running the extension.

  • Permissive policy: Follow that up with a secondary policy, or simply adjust the existing Default policy, that explicitly allows access to Claude for covered users.

The combined effect:

  • Extension users: Their traffic passes through the permissive policy, bypassing the DNS block. From there, the extension steps in and enforces the data-sharing requirement directly within the Claude interface.

  • Unprotected devices (fat clients): Without the extension in place, these devices hit the restrictive DNS policy first and are blocked outright: no access to Claude, no risk of sensitive data leaking out.
Secure Industries, Inc 101 Avenue of The Americas, Floor 9 New York, NY 10013