Summary
A Nature Medicine commentary describes clinical AI agents that use consistency-based gating to refer uncertain cases. It says the process after referral has not yet been tested.
A Nature Medicine News & Views article published on 2 October 2026 describes a study of on-premise AI agents for clinical decision-making. The commentary says the agents use consistency-based gating to refer uncertain cases, but that what happens after referral remains untested.
The associated open-access research article, “On-premise medical AI agents for reliable clinical decision-making,” was published on 15 September 2026. The commentary frames the work around two kinds of trust: operational trust in how an AI system functions within a clinical workflow, and decisional trust in its role in clinical decisions.
Referral is part of the safety question
An on-premise system runs on local computing infrastructure rather than depending solely on a cloud-hosted service. In the approach described, consistency-based gating is used to identify uncertain cases for referral instead of treating every case as one the agent should handle autonomously.
That creates a distinction between recognising uncertainty and resolving it. Escalation can provide a route for difficult cases, but its value in practice depends on what happens once a case is referred. The commentary identifies that downstream process as the missing link: it has not been tested in the study it discusses.
The finding is therefore about the design and limits of an AI-supported workflow, not evidence that referral improves patient outcomes. The commentary’s summary does not provide study methods or quantitative results, so the reported point is the use of uncertainty-based referral and the untested step that follows it.