We ran HealthBench on our health AI's safety layer. It scored lower than the bare model.

I build Tabibu, a health-information assistant. It adds a safety layer on top of a language model: emergency escalation, refusing prescription doses, answering from retrieved sources, and a reviewer t