A library patron assistant must refuse medical diagnosis requests and stay within library-service scope. Which control should be configured?

A library patron assistant should configure guardrails that block disallowed medical-diagnosis topics so the bot refuses diagnosis requests and stays within library-service scope. The control works because topic restrictions and safety policies can detect out-of-scope clinical asks, refuse diagnosis, and optionally redirect patrons to finding reputable information resources without practicing medicine. Scope guardrails protect residents and the library from inappropriate automated clinical advice while preserving legitimate catalog and reference help.

Guardrails that ground medical-diagnosis answers in library catalog holdings data fail because grounding still permits the assistant to provide diagnosis advice with citations rather than blocking the disallowed topic, leaving it to practice medicine outside library-service scope. Guardrails that mask medical-diagnosis entities in patron chat assistant responses fail because masking or redacting clinical terms only obscures output text and does not prevent the assistant from attempting medical diagnosis or operating outside library scope. Guardrails that filter profanity in medical-diagnosis replies from patron assistant fail because filtering profanity only removes offensive language and does not block disallowed medical-diagnosis topics, so the assistant can still generate out-of-scope clinical advice.

Exam caveat: distinguish “help me find consumer-health books or official public-health links” (often in-scope reference help) from “diagnose my symptoms” (out-of-scope); guardrails should encode that nuance. Operational check: implement topic blocks for diagnosis and treatment directives, test refusal quality on medical prompts, provide approved redirection language, and review mis-fire rates so legitimate library research questions about health topics are not over-blocked.