Microsoft Debuts Diagnostic AI System Aiming for Medical Superintelligence

The MAI-DxO system functions by orchestrating several AI agents that interview patients, order diagnostic tests, and debate clinical conclusions. In tests utilizing complex case studies from the New England Journal of Medicine, the software achieved an 83% accuracy rate, significantly outpacing the 36% success rate of general internists working with standard reference materials. Microsoft executives noted that while these results are promising, the system is designed to complement, rather than replace, specialist teams who typically manage such high-acuity cases.

Dominic King, vice president of Health at Microsoft AI, emphasized that the technology is intended to address the widening gap between the global demand for care and the availability of medical expertise. The research indicates that the system’s performance gains are driven by a combination of underlying model power—specifically with OpenAI’s GPT-5 and GPT-5.5—and Microsoft’s proprietary orchestration framework.

To move beyond simulations, Microsoft is launching real-world evaluations in partnership with Beth Israel Deaconess Medical Center in Boston. Clinicians will compare the AI’s diagnostic output against standard hospital workflows in both emergency and urgent care settings. Mustafa Suleyman, CEO of Microsoft AI, stated that the company is simultaneously collaborating with the Mayo Clinic to develop a dedicated foundation model for healthcare, with a broader release targeted for next year. The firm aims to leverage these tools to provide clinicians with predictive insights, effectively creating a digital flywheel to accelerate diagnostic accuracy and clinical decision support.