1. A security-conscious organization requires that their Copilot Studio agent NEVER generates responses from its language model using content from public internet websites, even if no SharePoint or internal knowledge source provides a relevant answer. The agent should only respond from explicitly configured internal knowledge sources or authored topics. Which combination of settings achieves this requirement? (Select TWO.)
- A. Disable the 'Allow the AI to use its own general knowledge' (model general knowledge) toggle in the generative AI settings✓ Correct
- B. Remove all public website URLs from the agent's knowledge sources✓ Correct
- C. Set the agent's topic fallback behavior to escalate to a human agent for all unmatched queries
- D. Enable the content moderation filter at the highest sensitivity level
- E. Turn off generative AI orchestration entirely so the agent only uses authored topics
Explanation
Disabling 'Allow the AI to use its own general knowledge' (A) prevents the language model from drawing on its parametric training data (which includes public internet content) when generating answers — grounding responses only in configured knowledge sources. Removing all public website URLs from knowledge sources (B) ensures the agent does not retrieve and use live public web content. Together these two settings confine the agent strictly to internal configured sources. Setting fallback to human escalation (C) only affects what happens when no topic matches; it does not prevent model-generated answers within matched topics or generative flows. The content moderation filter (D) filters harmful content but does not prevent the model from using general or public knowledge. Turning off generative AI orchestration entirely (E) would prevent knowledge-source-based answers as well as general knowledge answers — overcorrecting and preventing the agent from using the desired internal knowledge sources via generative answers.