A healthcare multi-agent triage helper must never recommend specific prescription dosages, even if a model would. Built-in generic safety filters are not domain-specific enough. What should you implement?
Select an answer to reveal the explanation.
Short Explanation
You need custom domain guardrails that encode the clinical policy, like banning specific prescription dosage recommendations, and you must validate them with synthetic clinical prompts before release. Generic filters aren't enough, and blunt controls like removing all medical knowledge or trusting users don't solve the problem.
Full Explanation
Implement custom domain guardrails encoding clinical policy, such as banning specific prescription dosage recommendations, and validate them with synthetic clinical prompts before release. Place checks on outputs and, where needed, on tools that could fetch dosing tables. Removing all medical knowledge sources is wrong: over-broad refusal destroys useful non-prescriptive guidance such as hours, locations, and general education, and is not precise policy engineering. Relying on users to ignore dosage advice is wrong because users cannot be the primary safety control. Increasing beam search width only is wrong because decoding parameters do not enforce clinical policy. Document guardrail ownership and update procedures when clinical policy changes.