Large language models in perioperative anesthesia: a risk-based roadmap for safe clinical adoption

Abstract

Perioperative uses are stratified by the consequence of failure: lower-risk information processing may be automated when outputs are traceable; intermediate-risk interpretive uses should augment clinicians and require explicit verification; high-risk or time-critical decisions should remain under direct specialist control. These tiers are not fixed properties of task categories; assignment depends on the intended use, downstream clinical consequences, degree of automation, and opportunity for independent verification