Artificial intelligence (AI) tools are more likely to provide incorrect medical advice when misinformation appears to come from an authoritative source, according to a new study published in The Lancet Digital Health.
Read More: How Artificial Intelligence Can Foster Peace and Interfaith Harmony?
Researchers tested 20 open-source and proprietary large language models (LLMs) and found that the software was more easily misled by errors embedded in realistic-looking doctors’ discharge summaries than by incorrect claims circulating on social media. The findings highlight growing concerns about the reliability of AI systems increasingly used in healthcare settings.
“Current AI systems can treat confident medical language as true by default, even when it’s clearly wrong,” said Dr. Eyal Klang of the Icahn School of Medicine at Mount Sinai in New York, who co-led the study. “For these models, what matters is less whether a claim is correct than how it is written.”
Researchers tested 9 leading LLMs with prompts containing medical misinformation. The result? Instead of correcting false premises, the AIs frequently validated them to satisfy the user, the AI is likely to act as a “yes man” and reinforce the lie than to push back pic.twitter.com/7ZcnqApAg2
— AI Agent Builder (@neuralagents_hq) February 10, 2026
To assess susceptibility to misinformation, the researchers exposed AI models to three types of content: real hospital discharge notes containing a single fabricated recommendation, common health myths drawn from Reddit, and 300 short clinical scenarios written by physicians. The models were then prompted with more than one million user-style questions and instructions related to the material.
Overall, the AI tools accepted and repeated fabricated information about 32% of the time. However, when misinformation appeared in what looked like an authentic hospital discharge note, the likelihood of AI believing and propagating it rose to nearly 47%, according to Dr. Girish Nadkarni, chief AI officer of the Mount Sinai Health System and a co-author of the study.
By contrast, AI systems were far more skeptical of social media content. When misinformation originated from Reddit posts, propagation dropped to just 9%.
The study also found that prompt phrasing influenced outcomes. Authoritative language, such as claims endorsed by a “senior clinician,” significantly increased the chances that AI would agree with false information.
Read More: Biden meets Microsoft, Google CEOs on AI dangers
Among the tested systems, OpenAI’s GPT models were the least susceptible to false claims, while some other models accepted up to 63.6% of incorrect statements. Researchers say the findings underscore the need for stronger safeguards before AI tools are fully embedded in clinical care.