AI Chatbots Misdiagnose Over 80 Percent of Early-Stage Medical Cases, Study Finds
A new study found that leading AI models from OpenAI and DeepSeek misdiagnose more than 80 percent of early-stage medical cases when given incomplete patient information, raising concerns about the reliability of AI tools in healthcare settings.
Leading AI models from OpenAI and DeepSeek misdiagnose more than 80 percent of early-stage medical cases when given incomplete patient information, according to a new study reported by Tech Startups in April 2026.
The findings expose significant limitations in current generative AI for healthcare. A separate Bloomberg-reported study found that AI chatbots gave misleading medical advice about half the time, highlighting reliability issues in sensitive consumer use cases.
The studies come as OpenAI is making a policy case for broader AI deployment in life sciences, arguing that AI could accelerate scientific discovery and treatment design. The company recently introduced GPT-5.4-Cyber, a defensive cybersecurity model, through a verified-access program.
Health experts said the misdiagnosis rates underscore the importance of using AI as a support tool rather than a replacement for clinical judgment. Patients who rely on AI chatbots for medical advice without consulting a doctor face real risks, particularly for conditions that require early intervention.
The U.S. Treasury is also seeking access to Anthropic's Mythos AI model to probe for vulnerabilities, treating advanced AI as a potential systemic risk for financial infrastructure. Anthropic's model reportedly uncovered thousands of software bugs faster than humans could fix them.
Google claims its AI helped block over 8.3 billion malicious ads in 2025, demonstrating the technology's effectiveness in some defensive applications. But the healthcare misdiagnosis data suggests the technology is not yet ready for high-stakes medical decisions without human oversight.
Psychiatrists are being urged to develop AI literacy skills, including training in intake, safety, ethics, and supervised AI integration, as AI tools become more common in mental health settings.
