Safety of AMIE for Emergency-Care Triage in Real-World Settings

Determine whether the Articulate Medical Intelligence Explorer (AMIE) conversational diagnostic AI system can safely identify and direct potentially medically unstable patients to emergency care during real-world patient interactions in ambulatory primary care settings.

Background

The study evaluated AMIE in pre-visit clinical conversations for urgent care but explicitly excluded patients requiring emergency care. While no safety stops were observed and conversational safety was favorable under physician oversight, the scenario of triaging potentially unstable patients to emergency settings was not encountered in the study population.

Because effective triage is a critical function of any intake or diagnostic system, the authors note that AMIE’s ability to safely manage situations necessitating emergency referral has not yet been empirically studied in real-world conditions.

References

Directing potentially medically unstable patients to emergency care is an important task in any triage system. No patients included in this study were triaged to emergency care settings and thus the ability of AMIE to safely navigate this scenario in the real world remains unstudied.

— A prospective clinical feasibility study of a conversational diagnostic AI in an ambulatory primary care clinic  (2603.08448 - Brodeur et al., 9 Mar 2026) in Discussion — Conversational Safety subsection