AI M&M Framework Proposed for Clinical AI Failure Review.

As clinical artificial intelligence becomes more integrated into healthcare, existing safety mechanisms are proving inadequate for understanding and learning from individual AI-related errors. Traditional aggregate model monitoring identifies performance shifts, and patient safety reporting captures adverse events, but neither explains the complex interplay between AI systems, clinicians, workflows, and institutional controls that leads to risk. To address this gap, researchers propose the AI Morbidity and Mortality (AI M&M) framework. This structured, blameless approach facilitates case-based review of clinical AI failures. It includes standardized case intake, evidence preservation, investigator-level reconstruction, tool-in-loop attribution, and corrective-action tracking. Events are classified across four dimensions: Trigger, Mechanism, Clinical Pathway, and Corrective Action, enabling a clear understanding of how vulnerabilities are exposed, risks produced, care impacted, and remediation assigned. The framework was demonstrated with five illustrative cases, showing high agreement among reviewers.