Microsoft tests multi-agent diagnostic AI in hospitals
Microsoft publishes results for its MAI-DxO diagnostic system and begins real-world hospital evaluation.
What happened
Microsoft reported that its MAI-DxO multi-agent diagnostic system correctly diagnosed 83% of simulated complex cases. Individually tested generalist physicians using reference materials reached 36%. The company is beginning a real-world evaluation with Beth Israel Deaconess Medical Center while participating patients continue to receive standard care.
Why it matters
A system that helps clinicians reason through difficult cases could improve access to expertise and support overstretched healthcare services. However, benchmark accuracy does not establish safer treatment or better patient outcomes. Prospective evaluation is essential before clinical deployment.
The bigger picture
Technology companies are moving medical AI beyond exam questions towards real workflows. The comparison must remain qualified: specialists and physician teams were not included, and the result came from a simulation. Claims of medical superintelligence are not established clinical facts.