Safety Evaluation Does Not Transfer: Domain-Specific Failure Patterns in Patient-Facing LLM Triage

Patient-facing large language models are increasingly used to guide care-seeking decisions, yet their safety is usually demonstrated in one clinical area and assumed to hold across others. Across 41 studies published 2023-2026 under a prospectively registered protocol, failures in physical-emergency studies concentrated toward both extremes of the urgency spectrum, with one stress-test under-triaging 51.6% of gold-standard emergencies. By contrast, suicide-risk studies located the most prominent failures among intermediate-risk cases. Failure location varied with clinical domain rather than remaining a stable property of the models, so a protocol validated in one domain cannot be assumed to detect the dominant failure in another. This is a cross-study comparison, so we also specify the matched experiment that would test transfer. Crisis-safeguard activation appeared sensitive to contextual reassurance independently of stated risk, and grounding in a structured triage protocol raised sensitivity for true emergencies from 0.73 to 0.98, beyond any between-model difference. With 27 of 41 studies using single-turn vignettes and none reporting input in any African language, the evidence base risks overstating safety and says nothing about African deployment settings. We set out what that means for evaluation there.