Story perspectives

AI Model Near 100% Accuracy Beats Doctors, Calls for Trials

5/1/2026

36 7

Summary

Manrai Model Excels

Researchers led by Manrai, co-senior author, tested a large language model on all medical-AI benchmarks and found it surpassed prior models and physician baselines.
Co-first author Peter Brodeur said the model scored near 100 percent on tests and identified likely diagnoses from raw emergency-department records, even with sparse early data.
Brodeur warned the model could suggest unnecessary testing that might harm patients, prompting the authors to demand rigorous prospective clinical trials.

Summary

Rodman Warns

Adam Rodman warned physicians could be personally liable for AI diagnostic errors treated like current decision-support tools.
Researchers noted the o1-preview model hallucinates and that its hallucination rate was not formally measured.
The model performed poorly on multimodal medical imaging benchmarks, limiting its usefulness for image or audio analysis.
Authors called for a “trust, but verify” stance and rigorous randomized controlled trials before wide clinical use.

Summary

Clinicians Warn

Billions are being invested in AI healthcare firms despite lacking a formal accountability framework.
UK clinicians have voiced concerns that AI errors could harm patients because liability rules are missing.
Researchers warn doctors may unconsciously defer to AI recommendations, which could erode judgment and let AI-doctor companies sideline physicians.
Patients still prefer human guidance for life-or-death decisions, prompting authors to call for a research agenda on safe AI tool use.

Summary

Reich Backs AI

David Reich, chief clinical officer for Mount Sinai Health System, said the AI is accurate, ready for prime time, and must be integrated into workflows.
Adam Rodman warned performance may drop with longer records because the study used brief emergency data.
Raj Manrai clarified the results do not justify replacing physicians.
The team urged forward-looking trials despite a challenging design.

History

Historical Context

The FDA’s April 18 2018 clearance of IDx-DR—developed by Dr. Michael Abràmoff’s Digital Diagnostics—marked the first autonomous AI diagnostic device, building on decades of CAD research that began with the 1998 FDA-approved mammography system. By 2025, 77 % of FDA-cleared AI tools target radiology, underscoring how early approvals paved the way for today’s emergency-department AI models.

1 of 5

Manrai Model Excels

Researchers led by Manrai, co-senior author, tested a large language model on all medical-AI benchmarks and found it surpassed prior models and physician baselines.
Co-first author Peter Brodeur said the model scored near 100 percent on tests and identified likely diagnoses from raw emergency-department records, even with sparse early data.
Brodeur warned the model could suggest unnecessary testing that might harm patients, prompting the authors to demand rigorous prospective clinical trials.

1 / 5