1. Tailspin Toys has a Copilot Studio copilot that uses generative answers over a Dataverse knowledge source. During testing, the team observes that the AI sometimes generates answers that mix data from unrelated Dataverse tables, causing confidentiality concerns. What is the BEST approach to address this?
- A. Disable generative answers entirely and author explicit topics for every supported question.
- B. Apply row-level security (RLS) on the Dataverse tables and ensure the copilot's service account only has access to permitted tables, combined with content moderation settings.✓ Correct
- C. Switch the knowledge source from Dataverse to a SharePoint document library to avoid cross-table data mixing.
- D. Set the generative answers node to require a confidence score above 90% before returning any response.
Explanation
The correct approach is to enforce row-level security (RLS) on the Dataverse tables so the copilot's service identity can only access permitted data, and to configure content moderation to further filter responses. This addresses both the data-access root cause and adds a safety layer. Option A is overly restrictive and eliminates valuable AI functionality. Option C does not inherently solve the problem—confidentiality issues can also occur with SharePoint if permissions are not managed. Option D is incorrect because the 'confidence score' threshold controls whether a generative answer is surfaced at all; it does not restrict WHICH Dataverse tables or rows are accessed during generation.