If you had itchy, sore, slightly darkened eyelids over the past year and turned to an AI chatbot for answers, you might have been told that you had a condition called bixonimania. There is just one problem — it doesn’t exist.
In a striking experiment, which is now raising fresh concerns about the reliability of artificial intelligence in healthcare, researchers deliberately invented a fake disease and watched as popular AI tools confidently diagnosed it, explained it, and even embedded it into scientific discourse.
The fictional condition was created in early 2024 by Swedish medical researcher Almira Osmanovic Thunström of the University of Gothenburg. She described bixonimania as a form of eyelid hyperpigmentation and irritation caused by blue light from screens with symptoms like pinkish and darkened lid, and itchy eyes. The goal was simple: to test whether large language models (LLMs) would absorb and reproduce fabricated medical misinformation.
To make the hoax convincing, Thunström went all in. She posted two preprints on SciProfiles, an academic social networking platform, under the fake author “Lazljiv Izgubljenovic” from the non-existent Asteria Horizon University, The papers included AI-generated profile photos, absurd funding sources like the “Professor Sideshow Bob Foundation”, and even acknowledgements to Starfleet Academy. One of the papers went as far as to openly admit the truth: “this entire paper is made up”, and that the study included fabricated patient data for “fifty made-up individuals.”
The signals were obvious, at least to a human reader. But they were not enough.
Ripple effects
Within weeks, by mid-April 2024, major AI chatbots began treating the condition as real. Microsoft’s Copilot described it as an “intriguing and relatively rare condition.” Google’s Gemini urged users to consult an ophthalmologist, Perplexity cited a bogus “prevalence of one in 90,000.” ChatGPT, meanwhile, blended the fictional illness seamlessly into legitimate advice on digital eye strain.
Trained on vast datasets scraped from the web, including repositories like Common Crawl, these systems appeared to treat the fabricated preprints as credible scientific material, ignoring the glaring red flags that would have tipped off human experts.
Even as late as March 2026, responses remained inconsistent. Some models hedged by calling it “not widely recognized”, while others affirmed it as a “proposed subtype of periorbital melanosis.”
The ripple effects extended beyond AI chatbots. A 2024 paper published in the Cureus journal by researchers from India’s Maharishi Markandeshwar Institute cited bixonimania as an “emerging periorbital melanosis linked to blue light.” The paper was eventually retracted in March 2026 for including “irrelevant references to a fictitious disease”.
Experts say the episode exposes a deeper vulnerability — not just in AI systems, but in the broader knowledge ecosystem increasingly shaped by them. As researchers and writers lean more heavily on AI–generated summaries and citations, the risk of unverified information slipping through grows significantly.
Also read: Aravallis were protected due to sacred belief. Now economic benefit has taken over
A broader concern
Part of the problem lies in how AI systems process information. In another study, published in The Lancet Digital Health in January 2026, it was found that LLMs are more likely to “hallucinate” or generate false information when content appears in a professional, medical format rather than informal sources like social media. Compounding the issue is inconsistency. The same AI tool could give contradictory answers depending on how a question is framed, sometimes treating bixonimania as a legitimate illness, and other times dismissing it as fictional.
For Thunström, the experiment came with its own ethical dilemmas. She deliberately chose a low-risk, non-life-threatening condition to minimise potential harm and consulted an ethics adviser before proceeding. “I wanted to make sure that we’re not creating more harm than good,” she said. The name itself, bixonimania, was chosen because it “sounded ridiculous”. “I wanted to be really clear to any physician or any medical staff that this is a made-up condition. No eye condition would be called mania–that’s a psychiatric term.”
Despite the absurdity, the fiction travelled, through algorithms, into answers, and even into published research. But the experiment underscores a broader concern — in high-stakes fields like healthcare, even brief windows of misinformation can have real-world consequences.
(Edited by Aamaan Alam Khan)