Microsoft's MAI-DxO Multi-Agent System Outperforms Generalist Doctors on Complex Diagnoses in Science Study
A study published in Science found that Microsoft's MAI-DxO, a multi-agent orchestration system acting like a panel of doctors, correctly diagnosed 83% of complex NEJM cases versus 36% for generalist physicians, with most gains attributable to underlying models like GPT-5 and GPT-5.5. Mustafa Suleyman framed the result as a step toward 'medical superintelligence' and announced a Mayo Clinic healthcare foundation model planned for next year. Microsoft will also run a large real-world evaluation with Beth Israel Deaconess Medical Center across emergency and urgent care settings.
Skynet Chance (0%): This is a domain-specific application with human-supervised clinical evaluation, so it adds little to loss-of-control risk, though it slightly normalizes delegating high-stakes decisions to AI agents.
Skynet Date (+0 days): The narrow medical focus and cautious, staged real-world testing do not meaningfully change the pace of general capability development or associated risks.
AGI Progress (0%): Frontier models reaching roughly 85% accuracy on complex diagnostic reasoning is evidence of strong expert-level reasoning, but the orchestration gains are modest and shrink with stronger models. Most of the improvement reflects already-known model scaling rather than a new capability.
AGI Date (+0 days): Microsoft's stated commitment to accelerate medical AI and to build real-world data flywheels slightly speeds expert-domain deployment. It does not substantially alter the overall AGI timeline.
<< All AI news for October 22, 2026
[ Get the daily index digest on Telegram → ]Related AI News
- OpenAI, PerturbAI, and Scripps Research Deploy Thousands of AI Agents to Mine Brain Cell Data 2026-10-21
- OpenAI Unveils Dots, Always-On Autonomous Agents Powered by GPT-6 Astra 2026-09-29
- Anthropic's AI-Driven Biology Lab Finds a New CRISPR-Like Enzyme System 2026-09-23
- Whistleblower Hotlines Launch for AI Agents Reporting on Misbehaving Peers 2026-09-15
- Anthropic Study Finds Claude Agents Sabotaging Each Other in Emergent Multi-Agent 'Turf Wars' 2026-08-13