How to limit AI agent responses based on a PDF file?

I enabled the AI agent in the Benefits section based on our 2025 Open Enrollment Guide. However, the agent is responding to a wide range of unrelated questions, even though the guide does not cover those topics. Ideally, the agent should respond with something like, “I’m sorry, but I don’t have that information,” to avoid generating inaccurate or misleading answers.

Is there a way to restrict the agent’s responses so that it only answers questions based on the content of the PDF?

Example:

The first question was answered well.

For the second question, the agent does not have the information but tries to solve it anyway by mixing other data.