Microsoft's AI diagnostic system just beat generalist doctors by more than two to one on complex medical cases in a peer-reviewed study — and the race to collapse the medical establishment's diagnostic monopoly is on. The question for working Americans isn't whether the machine can do it. It's who owns the diagnosis — and the data — when Big Tech and Big Medicine merge.

Published Thursday in Science, the study found Microsoft's Diagnostic Orchestrator — dubbed MAI-DxO — correctly diagnosed 83% of complex cases pulled from the New England Journal of Medicine. Primary care doctors and general internists working with reference materials got 36%. On emergency department cases from Beth Israel Deaconess Medical Center in Boston, the AI hit 80% accuracy versus 47% for human physicians. Microsoft AI CEO Mustafa Suleyman called it a step toward "medical superintelligence."

Here's the catch both outlets soft-pedaled: those complex NEJM cases are normally handled by specialists or full physician teams — and neither specialists nor teams were part of the comparison. The AI beat solo generalists. That's not nothing, but it's not the whole story either. TechCrunch noted this caveat; The Verge buried it.

The system works by coordinating multiple AI agents that question a patient, order tests, and debate a diagnosis — essentially simulating a panel of doctors. Microsoft compiled the test cases into a benchmark called SDBench, which it plans to open-source.

Follow the money. The best results came running DxO on OpenAI's GPT-5 and GPT-4.1. On its own, GPT-5 scored 78.9% on complex cases at an estimated $8,400 per case. DxO cut about $3,000 from that and added roughly four points of accuracy. With the newer GPT-5.5, DxO pushed accuracy from 72% to nearly 87% on complex cases — but gains vanished on emergency department cases, where both scored about 85% at similar cost. The orchestration layer helps most when the underlying model is weaker. As models improve, DxO's value-add shrinks. Microsoft's own researchers admitted accuracy gains naturally contract as models approach their ceiling.

Microsoft is already moving to lock up the data pipeline. Suleyman told TechCrunch the company is working with the Mayo Clinic to train a healthcare foundation model for release next year. "I really believe that in 2027, we will be on the verge of a coding style breakthrough for medical records," Suleyman said. He predicted the model will "stream highly accurate predictions about how a medical record might unfold."

That's the real headline. Microsoft isn't just building a diagnostic tool — it's positioning itself as the infrastructure layer for American healthcare. Mayo Clinic supplies the institutional credibility and the patient data; Microsoft supplies the compute and the platform. When your medical records stream through a Microsoft foundation model, who controls access? Who sets the price? Who decides when the AI's diagnosis overrides the doctor's — or your own judgment?

Suleyman conceded the system "needs clinical validation" and has only been tested on historical cases. The Verge reported that Microsoft outlines six areas of healthcare AI mastery it's pursuing, including treatment decisions and coordinating clinical care. This isn't a diagnostic assistant. It's a bid for the operating system of medicine.

Dominic King, Microsoft AI's vice president of Health, framed the stakes as a supply-demand mismatch: "The biggest challenge in healthcare delivery at the moment is the growing mismatch between demand and supply of medical expertise." That's true. It's also the sales pitch for replacing that expertise with a product Microsoft licenses.

The medical establishment has earned the public's skepticism — opaque pricing, gatekept care, bureaucratic bloat. AI that democratizes diagnostic knowledge could be a genuine liberation. But liberation and monopoly look the same at the product launch. The difference is who holds the keys when the demo ends.