Autonomous medical AI outperforms doctors in simulated EHR cases

A study published in *Nature* has introduced MIRA, an autonomous AI agent that operates within a sandboxed electronic health record (EHR) environment. The system was evaluated in simulated emergency cases and was reported to outperform human physicians in diagnostic accuracy and treatment decisions. The research highlights the potential for AI to assist in clinical decision-making, though the simulations were conducted in a controlled, sandboxed setting rather than in real clinical practice.

The study did not involve live patient interactions; all cases were synthetic or de-identified data presented to the AI through a simulated EHR interface. MIRA autonomously reviewed patient histories, lab results, and other clinical data to generate diagnoses and management plans. The performance metrics compared MIRA's outputs against those of physicians who were given the same simulated cases. The authors note that the sandbox environment removes real-world constraints such as time pressure and incomplete information, which may affect generalisability.

While the results suggest that autonomous AI systems could eventually support clinical workflows, the authors caution that real-world deployment would require rigorous validation, including prospective trials and regulatory approval. The *Nature* publication has generated discussion among medical informatics researchers about the appropriate benchmarks for evaluating AI in healthcare. Further studies are expected to test MIRA in more realistic settings that incorporate patient communication and dynamic clinical environments.