A group of researchers invented an eye disease to see if AI would catch it. It didn't. Instead, ChatGPT, Gemini, Copilot and Perplexity all started explaining the fake condition to users as if it were real medicine.
The disease was called "bixonimania," a supposed eye disorder caused by too much blue light. It doesn't exist. Almira Osmanovic Thunström and her team made it up, wrote two fake preprints, and uploaded them to Medium and an academic social network in 2024 to see what would happen once the papers entered the pipeline of data that chatbots feed on.
To make the test fair, they left obvious clues. The lead "author" was named Lazljiv Izgubljenovic, which translates roughly to "lying loser" in Bosnian. The affiliations included Starfleet Academy and the USS Enterprise. The funding section thanked the Professor Sideshow Bob Foundation. One paper even included the line "this entire paper is made up." The suffix "-mania" doesn't belong on an eye condition either. It's used for psychiatric disorders. A first-year med student would notice.
The chatbots did not.
- Microsoft Copilot called bixonimania "an intriguing and relatively rare condition."
- Google Gemini confirmed it was caused by blue light and told users to see an ophthalmologist.
- Perplexity went further and invented a prevalence rate of 1 in 90,000, a number that appears nowhere in the source material because the source material doesn't exist.
- ChatGPT did the strangest thing. Some versions started calling it a "proposed new subtype," while other responses from the same platform, days apart, dismissed it as "probably a made-up, fringe, or pseudoscientific label."
"I really wanted to have a clear case that leaves breadcrumbs throughout the whole system," Thunström told Smithsonian Magazine, describing why she buried so many tells in the papers. The breadcrumbs were the whole point. The chatbots walked past every one of them.
The uncomfortable part is what these same companies say about themselves. In a policy statement this summer, the FTC quoted the AI labs' own marketing pages. OpenAI pitches ChatGPT for Healthcare as pulling from "millions of peer-reviewed studies, clinical guidelines, and public health sources" to give "clinical answers with citations." Anthropic describes Claude as talking to you like "a brilliant friend" who provides "information grounded in current, reliable evidence." X.ai calls Grok "your truth-seeking AI companion for unfiltered answers." None of that survived contact with a fake eye disease and a paper credited to Sideshow Bob.
Alex Ruani, a misinformation researcher at University College London, called the experiment "a master class on how mis- and disinformation operates," telling Smithsonian Magazine that when the scientific pipeline fails to filter out compromised content early, everything downstream inherits the problem. That includes the models trained on it, and the users asking those models about their health.
We've seen a security version of this play out with AI-invented software packages that hackers eventually registered themselves. This is the medical version, and it lands differently, because a lot more people ask a chatbot about their eyes than about their code.
The interesting thing about bixonimania isn't that the chatbots got it wrong. It's that they got it wrong with confidence, with citations, and in exactly the tone their marketing pages promise. The models were doing the job they were trained to do, which is to sound authoritative about whatever shows up in their training data. Nobody built a filter for "is this actually real." Until someone does, every clinical answer from a chatbot is really a well-phrased guess about what a clinical answer sounds like. The labs selling these tools as medical companions might want to sit with that one for a while.
