Answers
Refusals and escalations
When AI should refuse or escalate instead of guessing.
5 min read · Updated Jul 2026
Applications should refuse or escalate when knowledge is missing, policy blocks retrieval, or questions need human agents.
What you will learn
- Policy decisions leading to refusal
- Configuring escalation messages
- Testing refusal paths in Playground
- Avoiding confident guesses
Steps
- 1
Review policy rules
Permissions, open Support Assistant policy escalation section.
- 2
Define refusal topics
Add rules for legal threats and employee data requests.
- 3
Set escalation copy
Provide message directing users to Nintendo support contact channel.
- 4
Test refusal
Playground: Ask for another user account details.
- 5
Confirm no citation leak
Ensure response refuses without internal snippets.
- 6
Save refusal test case
Add to evaluation set for regressions.
Common issues
- Model guesses instead of refusing.
- Tighten unavailable rules and remove borderline docs from scope.
- Over-refusal on valid questions.
- Verify documents exist and policy allows collection.
Ready to try it in Wiki?
Start with a Wiki account and set up your Nintendo workspace today.
Was this helpful?