A healthcare multi-agent system must never recommend changing prescribed medications and must refuse diagnosis claims. Built-in generic safety filters help but do not encode this clinical policy. What should you add?
Select an answer to reveal the explanation.
Short Explanation
You need custom domain-specific guardrails, not generic filters. They must catch medication-change and diagnosis intents in inputs and outputs, block or rewrite violations, and log events for compliance. Manual fax retyping isn't a guardrail architecture.
Full Explanation
Correct answer — Custom domain-specific guardrails (and policy checks) that detect medication-change or diagnosis intents in user inputs and agent outputs, block or rewrite violations, and log policy events for compliance.
You must design custom guardrails for domain-specific constraints. Healthcare medication and diagnosis policies need dedicated detection and enforcement on inputs/outputs with compliance logging—not only generic toxicity filters.
Why the generic open-web toxicity list is wrong: it does not encode clinical non-diagnosis or non-prescribing rules.
Why nurse fax retyping is wrong: manual fax retyping is not a scalable technical guardrail strategy.
Why unrestricted clinical advice is wrong: it violates clinical policy and responsible AI constraints.
Exam tip: Domain policy → custom guardrails + audit, not only stock filters.