An AI chatbot should never answer questions it is not authorized or qualified to answer. That can include legal, medical, financial, emergency, account-security, or confidential business questions.
In practice
It should also avoid inventing discounts, guarantees, policies, timelines, product capabilities, or availability. If the answer is not in approved business information, escalation is usually better.
What to watch
Clear refusal and handoff rules protect both the visitor and the business. The bot should be helpful without pretending to have authority it does not have.