Yes, and it should. A well-designed internal AI assistant should recognize when it lacks sufficient evidence, avoid guessing and direct employees to the appropriate person, process or source document. Refusing to answer is often a sign of a mature knowledge system, not a weak one.
The Biggest Risk Is Not “I Don’t Know”
Employees lose confidence much faster when the assistant confidently gives the wrong answer. For a purchase approval, leave rule or safety question, an honest gap is safer than an invented policy.
Why AI Sometimes Shouldn’t Answer
The information may not exist, documents may conflict, the employee may lack permission or the question may require human judgment. Microsoft’s Copilot Studio generative answers FAQ explains why grounding and testing remain necessary.
Low Confidence Is a Feature
Useful responses include: “I couldn’t find an approved source,” “Multiple documents disagree,” “This appears outdated,” or “This requires human approval.”
Every Unanswered Question Improves the Hub
Repeated gaps reveal missing policies, undocumented exceptions and weak source ownership. They should become a documentation backlog rather than disappear inside chat logs.
Escalation Should Be Intentional
The assistant should identify the owner, policy, help desk, workflow or manager responsible for the next step. Microsoft’s NIST AI Risk Management Framework emphasizes governance, measurement and risk treatment rather than blind automation.
Where Maisy Fits
The AskMaisy implementation guide explains why ownership and authoritative sources matter. The security and architecture guide explains refusal, escalation and permission-aware retrieval.
The Bottom Line
A trustworthy assistant distinguishes what the organization knows, what it does not know and what requires human judgment.
