Do You Have Bixonimania?
AI medical information can be flawed — or even doctored.
Posted May 15, 2026 | Reviewed by Monica Vilhauer Ph.D.
Are your eyes itchy? A little sore? Are your eyelids reddish? (Stokel-Walker, 2026). If you looked those symptoms up on AI earlier this year, you may have been told you have bixonimania. Except you most certainly don’t: Bixonimania is a fake condition made up by Almira Osmanovic Thunström, a medical researcher in Sweden. In 2024, she uploaded two fake preprints to an online platform. Preprints are reports of studies that haven’t yet been peer-reviewed. They are a way to help researchers know what results others in their field are finding before the studies go through the often-long process that leads to publication in a peer-reviewed journal article.
Osmanovic Thunström was not trying to be malicious. Rather, she wanted to see if it was possible to create misinformation that is as potentially damaging as a completely invented disease. She picked the name bixonimania because it “sounded ridiculous.” She wanted her deception to be clear to anyone with medical training “because no eye condition would be called mania — that’s a psychiatric term.” She also inserted plenty of clues into the actual preprints to make sure readers would know it was fake. She reported funding from “the Professor Sideshow Bob Foundation for its work in advanced trickery” and used a fictional researcher name, university, and city. Toward the beginning she even wrote: “This entire paper is made up.”
Within weeks of uploading the preprints, AI was already including the condition in their responses, even though it didn’t exist. Perhaps even worse, the fake preprints started showing up in actual peer-reviewed journal articles, likely because of researchers relying on AI sources without checking them. We’ve already written about the wholesale invention of sources by AI . Osmanovic Thunström has now demonstrated that this tendency toward source hallucination could be compounded by malicious actors who seed preprint servers with fake papers. (Note: Since this particular deception has been written about in the news, AI is no longer fooled by bixonimania.) These disturbing trends are occurring at a critical time when accurate health data are being removed from U.S. government websites.
Dr. AI Is Often Wrong
How bad is the problem of AI inaccuracy in medicine? Researchers recently asked five AI chatbots 10 questions from a range of medical areas (Tiller et al., 2026). The chatbots produced “problematic” responses about half the time, and about 20% of all responses were “highly problematic.” For example, the researchers reported responses that exaggerated the risks of vaccines that are well established as safe, suggested unsupported alternative treatments for cancer (e.g., herbal remedies), or provided inaccurate information regarding insurance coverage. The researchers concluded: “Continued deployment without public education and oversight risks amplifying misinformation.”
These statistics are alarming, especially when you think about the actions people might take based on “problematic” information. Indeed, Joe Riley likely died because of his reliance on AI. The New York Times recently reported the story of Joe and his son, Ben Riley. In 2025, Ben discovered that his dad had been ignoring advice from his doctor to start cancer treatment in favor of advice from AI that led him to forego treatment altogether. Ironically, Joe, a retired neuroscientist , had a background in science. Also ironically, his own son Ben had founded a nonprofit “to train teachers in cognitive science to better understand how their students thought and learned.” You would think this combination would overcome misinformation: a son focused on helping people understand thinking could convince a scientist dad to think critically about information from generative AI .
Tragically, neither Ben nor Joe’s doctor could convince him in time. Joe was sure, despite all evidence to the contrary, that he had a rare cancer complication that treatment would worsen. By the time Joe agreed to start treatment, a year after it was first recommended, it was too late to save him. He died in December, 2025.
We’re Just Not Good at Prompting AI
A recent study suggested that Joe is not alone in his failure to elicit accurate information from AI (Bean et al., 2026). The researchers found that AI was fairly accurate in generating correct diagnoses when given information about a medical condition generated by a physician. But diagnostic accuracy rates dropped precipitously — from about 95% to about 35% — when non-physician research participants interacted with the same AI models. Why? The general public is more likely to write unclear prompts and to provide incomplete information.
This seems likely in Joe’s case. He had been specifically asking AI about the unusual complication along with a listing of his own symptoms, perhaps inadvertently prompting the AI to incorrectly suggest that he had it. When his son Ben showed Joe’s AI output to three experts on this complication, they all agreed Joe had been misled. One expert said that he couldn’t recognize his own research in the AI summary of it.
Joe’s son Ben is clear that he doesn’t think AI killed his father. He noted that Joe had always had a strong mistrust of doctors. Yet, AI-generated misinformation fueled Joe’s skepticism which couldn’t be overcome by his background in science or by a family and doctor who tried their best to persuade him. This is why Ben is speaking out. In his words, “There’s nothing I can do to change the past, of course. But I can for damn sure keep working to raise the consciousness of others.”
Bean, A. M., Payne, R. E., Parsons, G., Kirk, H. W., Ciro, J., Mosquera-Gómez, R., Hincapié M, S., Ekanayaka, A. S., Tarassenko. L., Rocher, L., & Mahd, A. (2026). Reliability of LLMs as medical assistants for the general public: A randomized preregistered study. Nature Medicine, 32 , 609–615. https://doi.org/10.1038/s41591-025-04074-y
Stokel-Walker, C. (2026, April 7). Scientists invented a fake disease. AI told people it was real. Nature. https://www.nature.com/articles/d41586-026-01100-y
Tiller, N. B., Marcon, A. R., Zenone, M. , Kidd, K. E., Jeukendrup, A. E., Master, Z., & Caulfield, T. (2026). Generative artificial intelligence-driven chatbots and medical misinformation: An accuracy, referencing and readability audit. BMJ Open, 16 :e112695. https://doi.org/10.1136/bmjopen-2025-112695
Share this post Facebook Bluesky Linkedin Email
There was a problem adding your email address. Please try again.
By submitting your information you agree to the Psychology Today Terms & Conditions and Privacy Policy
Susan A. Nolan, Ph.D. , is a professor of psychology at Seton Hall University and the author of textbooks on statistics and psychology.
Michael Kimball is the author of eight books, the host of an NBA podcast, and an editor of textbooks on statistics and psychology.
Get the help you need from a therapist near you–a FREE service from Psychology Today.
This article is part of the Bringwise Psychology Journal — daily insights on human behavior, mental health, and personal growth.