A new scientific report from the University of Oxford, published on February 9, 2026, has highlighted significant limitations of Large Language Models (LLMs) in providing health diagnostic support. The study warns users to exercise extreme caution when seeking medical guidance from AI tools.
The research involved nearly 1,300 participants across 10 diverse medical scenarios. Participants were randomly assigned to use AI chatbots—including GPT-4o, Llama 3, and Command R+—or traditional internet search engines. The findings revealed:
-
Low accuracy rates: Only 30% of the AI group received an accurate diagnosis, and only 45% were provided with appropriate treatment directions.
-
No advantage over search engines: In real-world applications, AI performance showed no significant improvement compared to standard internet searches.
-
Communication barriers: While AI often scores highly in theoretical medical exams, practical challenges arise because users frequently fail to provide comprehensive clinical details or struggle to interpret complex AI-generated advice.
Experts warn that AI may overlook critical emergency warning signs, potentially causing patients to delay seeking professional medical intervention. Consequently, consulting official healthcare sources and undergoing in-person examinations remains the most reliable method for health protection.