AI Models Misdiagnose Over 80 Percent of Early-Stage Medical Cases with Incomplete Data

A new study found that leading AI models from OpenAI and DeepSeek misdiagnose more than 80 percent of early-stage medical cases when provided with incomplete patient information.

The research, published in April 2026, tested the models on scenarios where key details about a patient's history, symptoms, or test results were missing. The high error rate highlights a significant limitation of current AI systems in real-world clinical settings, where incomplete information is common.

Researchers called for improved training methods that prepare AI models for partial-data scenarios. They also recommended hybrid human-AI systems, where AI assists clinicians rather than replacing them, and stricter regulatory frameworks for AI tools used in healthcare.

The findings come as AI companies push deeper into medical applications. Several startups are developing AI diagnostic tools, and major health systems are piloting AI for tasks ranging from reading medical images to triaging patients.

Proponents of AI in healthcare argue that even imperfect AI can improve outcomes by catching cases that human clinicians might miss. Critics say the high error rate in this study shows the technology is not ready for clinical use without significant human oversight.

The study also noted that AI models performed better when given complete information, suggesting that improving data collection and sharing in healthcare could help close the gap.

Regulatory agencies in the United States and Europe are developing frameworks for evaluating AI medical devices, but formal standards for AI diagnostic tools remain limited.