Human-in-the-Loop Approval Settings

Ensure your LLM has Approvals Required enabled to prevent unauthorized or accidental modifications to your security environment. When Approval is configured in your LLM Connector settings, each supported AI application will display an approval prompt as follows:

  • Claude: When Claude attempts to call a write tool (e.g., creating or updating a policy), it displays a prompt requiring you to click Allow once or Allow for this session.

  • ChatGPT: Prompt cards pop up during action workflows asking for explicit confirmation before committing changes to Forcepoint.

  • Microsoft CoPilot: Displays an action approval prompt detailing the intended payload before execution.

No changes are committed to the Forcepoint platform until explicit human confirmation is received in the client UI.