If your eyes have been itchy, sore, or slightly discolored lately after staring at your laptop, you might have done what millions of people do every day. You probably opened up an AI chatbot and typed in your symptoms.
If you did this, there's a decent chance a major artificial intelligence model told you that you were suffering from a rare, emerging condition called bixonimania. It explained that the illness is caused by blue light exposure and advised you to see an ophthalmologist immediately.
There's just one problem. Bixonimania doesn't exist.
It's a completely fabricated illness invented by a Swedish medical researcher named Almira Osmanovic Thunström to see if popular AI models could spot a blatant lie. They couldn't. Instead, they swallowed the fake science whole and started diagnosing real people with it.
This bizarre experiment exposes a terrifying truth about the tech we rely on for health answers. AI chatbots aren't searching for truth. They're just matching patterns, and they'll happily repeat dangerous misinformation if it's dressed up in the right vocabulary.
The Anatomy of a Brilliant Medical Hoax
Osmanovic Thunström, a researcher at the University of Gothenburg, didn't try to hide her deception. She didn't write a deeply convincing, flawless paper to trick the medical community. She did the exact opposite.
In early 2024, she uploaded two fake preprints—academic papers that haven't been peer-reviewed yet—to an online repository. She intentionally stuffed these papers with ridiculous red flags that any human with a brain would catch instantly.
First, she named the disease bixonimania. In medical terminology, the suffix "-mania" is used strictly in psychiatry to denote mental states, not physical eye disorders. No eye doctor would ever name a condition this way.
Then she created a fictitious lead author named Lazljiv Izgubljenovic, a Serbian name that translates directly to "lying loser". She generated a fake, AI-made photo for his profile.
If that wasn't enough, the papers listed the author's institution as Asteria Horizon University in Nova City, California—places that do not exist on any map.
But the real comedy was hidden in the funding and acknowledgements sections. The paper explicitly thanked "Professor Maria Bohm at The Starfleet Academy" for her work onboard the USS Enterprise. It stated that funding came from the "Professor Sideshow Bob Foundation for its work in advanced trickery" alongside the "University of Fellowship of the Ring and the Galactic Triad".
At one point, the text flat-out stated: "This entire paper is made up".
It was a total joke. Yet, the internet's most celebrated technology fell for it completely.
How the Chatbots Failed the Test
Within weeks of the fake papers hitting the web, major AI platforms indexed the data and integrated it into their knowledge bases. They completely missed the references to Star Trek, Lord of the Rings, and Sideshow Bob.
When users asked about these specific eye symptoms or mentioned the name of the disease, the bots answered with absolute authority.
Microsoft Copilot told users that "Bixonimania is indeed an intriguing and relatively rare condition".
Google Gemini was just as confident, stating that "Bixonimania is a condition caused by excessive exposure to blue light" and urging users to see a specialist.
Perplexity AI went a step further in its hallucination, calculating a fictional epidemiology and asserting that the disorder affected precisely one in 90,000 individuals globally.
OpenAI's ChatGPT actively diagnosed users, walking them through whether their specific screen habits meant they had contracted this nonexistent disease.
Think about that. Millions of people use these tools for medical advice every day. If an AI can be tricked into validating a disease funded by a cartoon villain, what else is it getting wrong?
The tech companies often hide behind the defense that their models are getting smarter. An OpenAI spokesperson noted that newer versions of ChatGPT are better at handling medical information than they were when the study began. But updates are a game of whack-a-mole. While the bots have since been manually corrected to recognize bixonimania, the underlying flaw that allowed the error to happen remains untouched.
Why AI Chatbots Make Dangerous Doctors
The problem isn't that AI is stupid. The problem is that AI doesn't understand what "truth" means.
Large language models operate on probability, not fact-checking. They look at a prompt, look at their training data, and predict which words should follow each other to form a convincing answer. They're text-generation engines, not truth engines.
Separate research into medical AI systems found a fascinating and terrifying quirk: chatbots are much more likely to hallucinate and believe false information if the text is written in formal, authoritative medical language.
If a piece of misinformation is posted on a casual social media forum, the AI might ignore it or treat it with skepticism. But if you format that exact same lie like a clinical trial report or a hospital discharge note, the model's confidence scores skyrocket.
Because Osmanovic Thunström's fake papers used structured academic language, the AI training pipelines prioritized them. The systems saw words like "periorbital hyperpigmentation" and assumed it was high-quality medical science, entirely blind to the fact that the text explicitly called itself a fabrication.
The Contamination of Real Medical Science
The story gets worse. This misinformation didn't just stay inside the chatbot sandboxes. It actively contaminated real scientific literature.
Human medical researchers, likely using AI tools to help write their own papers, started citing the fake bixonimania preprints without actually reading them. A peer-reviewed medical journal called Cureus, published by the prestigious Springer-Nature group, actually printed a paper that cited bixonimania as a real, emerging eye disorder. Editors had to issue a quiet retraction once the hoax came to light.
This isn't an isolated incident. An audit of 2.5 million biomedical papers revealed that roughly 1 in 277 papers published contained fabricated or hallucinated references. That's a massive, twelve-fold increase compared to just a few years ago.
We're entering an era where AI is feeding on fake data, generating bad science, and then real scientists are using AI to write papers that cite that bad science. It's a feedback loop of garbage.
How to Protect Yourself From Dr. AI
You shouldn't rely on a chatbot to tell you if you're sick. It's that simple. If you must use tech to screen your health symptoms before making a doctor's appointment, change how you do it.
- Stop using general conversational chatbots for diagnosis. Tools like ChatGPT, Gemini, and Copilot are built for general writing and coding. They are not calibrated to guarantee medical accuracy.
- Stick to dedicated, verified medical databases. If you want to look up symptoms online, use platforms with strict editorial oversight like Mayo Clinic, Johns Hopkins Medicine, or WebMD. They might feel old-fashioned compared to a chat interface, but they don't get their funding from Starfleet Academy.
- Look for manual source verification. If an AI tool gives you a health claim, demand the exact source. If it can't provide a link to a verified, peer-reviewed study in a major journal, ignore the advice entirely.
- Talk to a human practitioner. A chatbot cannot look at your eyes, run a blood test, or understand your personal medical history. It can only guess what words you want to read.
Tech companies want us to believe these models are ready to replace human experts. The bixonimania experiment proved they aren't even close. Treat AI medical advice with extreme skepticism, because the bot diagnosing your symptoms might just be playing a joke it doesn't even understand.
