Story perspectives

OpenAI's o1 Beats Doctors in Diagnostic Test, Raises Safety Concerns

6/18/2026

Summary

OpenAI Beats Doctors

U.S. researchers published a study on April 30, 2026 in Science showing OpenAI’s o1 preview outperformed physicians in a diagnostic obstacle course.
Adam Rodman and Arjun Manrai presented the findings at a press conference before publication.
Beth Israel Deaconess Medical Center began offering an AI-powered clinical reasoning tool to clinicians without FDA approval.
Eric Topol warned that text-only large language models miss non-verbal cues and image review, limiting real-world use.

Summary

Hospitals Face Blame

A npj Digital Public Health experiment found the public blames hospitals more when AI alone or with minimal physician input causes a missed pneumonia diagnosis.
The result derives from two online studies with 299 and 602 participants evaluating the hypothetical misdiagnosis.
Both “autonomous” AI and “sequential” AI-physician collaboration made participants rate hospitals as unfavorably as AI-only, while “interactive” collaboration—radiologists reviewing AI-flagged areas and the full image—lowered blame and complaint intent.

1 of 2

OpenAI Beats Doctors

U.S. researchers published a study on April 30, 2026 in Science showing OpenAI’s o1 preview outperformed physicians in a diagnostic obstacle course.
Adam Rodman and Arjun Manrai presented the findings at a press conference before publication.
Beth Israel Deaconess Medical Center began offering an AI-powered clinical reasoning tool to clinicians without FDA approval.
Eric Topol warned that text-only large language models miss non-verbal cues and image review, limiting real-world use.

1 / 2