AI Chatbots Spread Medical Misinformation About Fabricated Disease Bixonimania
Researchers at Sweden's University of Gothenburg intentionally created a fictional eye disease called "bixonimania" to test how large language models handle false medical information. Multiple AI chatbots including ChatGPT, Gemini, Copilot, and Perplexity began treating the fabricated disease as real and providing medical advice about it, and the hoax eventually led to citation in a peer-reviewed medical journal.
What happened
In March 2024, Almira Osmanovic Thunström of the University of Gothenburg and her team conducted an experiment to demonstrate how large language models ingest and spread false information by creating a fictional eye disease called 'bixonimania.' Two blog posts describing the fabricated disease were published on Medium on March 15, 2024, followed by two fake academic preprints uploaded to SciProfiles on April 26 and May 6, 2024, attributed to a non-existent author named 'Lazljiv Izgubljenovic' at a fictional university in a fictional California city, with obvious markers indicating the papers were fake including references to science fiction and cartoon institutions.
Within weeks of these posts appearing online, multiple deployed AI chatbots began treating bixonimania as a real disease and providing medical guidance about it. Microsoft Copilot described it as a 'confirmed and rare disease,' Google Gemini claimed it was caused by excessive blue light exposure and recommended seeing an eye doctor, Perplexity AI provided a specific prevalence rate of approximately one in 90,000 people, and ChatGPT would classify user symptoms as potentially matching the fabricated condition.
The fabricated disease was later cited in a peer-reviewed medical journal: in March 2026, a paper published in the journal Cureus by researchers at Maharishi Markandeshwar Medical College referenced one of the fake preprints and described bixonimania as an emerging form of periorbital melanosis related to blue light exposure. After Nature journalists inquired about the paper, Cureus retracted it on March 30, 2026, stating the article contained unrelated references including a citation to the fabricated disease.
Reported impact
The experiment demonstrated that AI chatbots can confidently provide false medical information to users seeking health advice. Researchers noted the experiment 'succeeded too well,' as models continued to reference and build upon the false information despite obvious indicators it was fabricated.
The hoax exposed broader vulnerabilities in scientific publishing and information verification systems, as false information created intentionally as an educational experiment eventually penetrated peer-reviewed medical literature, suggesting that some researchers may be relying on AI-generated content rather than reading original papers carefully.
Index analysis
Multiple AI systems hallucinated medical authority about a non-existent disease, creating plausible-sounding clinical descriptions. The systems varied in their responses, sometimes claiming uncertainty and other times providing definitive diagnoses depending on how the question was phrased.
The vulnerability was amplified by the professional formatting of the fake papers, which mimicked legitimate academic and clinical documentation; when AI input was presented in the format of discharge summaries or clinical papers rather than informal social media language, models were more prone to generating false information by building on existing text.
Governance controls
When questioned about their responses, the AI companies acknowledged problems with earlier model versions but claimed more recent versions had improved: Perplexity stated it prioritizes accuracy, OpenAI claimed significant improvements in medical information safety in current ChatGPT models, and Google said early models had limitations they were working to address. Microsoft did not respond to requests for comment.
The experiment's creator, Almira Osmanovic Thunström, consulted with ethics advisors before conducting the study and deliberately chose a low-risk topic (minor skin issues) to minimize potential harm from introducing false medical information into scientific literature. She noted concerns that seeding a fake disease in scientific literature could cause real-world harm.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.