What happened
Stanford HAI published a policy brief introducing the Ensemble Monitoring Model (EMM), a framework for real-time monitoring of radiological AI tools. Radiological AI accounts for approximately 76% of FDA-authorized AI-enabled medical devices as of December 2025, yet most deployed systems lack robust performance monitoring. EMM measures agreement between a primary AI model and an ensemble of five independent submodels to estimate uncertainty without requiring access to black box model components, inspired by clinical consensus practices. Using a large dataset focused on brain bleed detection, the framework demonstrates that EMM can reduce radiologists' cognitive burden by characterizing AI model uncertainty in real time at the point of care and guiding appropriate responses when cases are flagged for reduced accuracy. The brief argues that continuous performance monitoring should be treated as a core component of responsible AI deployment in healthcare.
Why it matters
Healthcare AI governance cannot stop at product approval. This framework provides a practical, deployable method for post-market surveillance of clinical AI systems that does not require vendor cooperation or access to proprietary model internals—a critical enabler for hospitals and health systems that must oversee third-party AI tools. Policymakers should view methods like EMM as a component of total lifecycle AI regulation.
Action needed
Hospital CIOs and clinical informatics leaders should pilot EMM-style uncertainty monitoring on at least one radiological AI system by Q4 2026 and brief the board on post-deployment oversight gaps.