Scientists invented a fake disease. AI told people it was real

Got sore, itchy eyes? Type your symptoms into a chatbot and you might get an answer no doctor ever taught you: bixonimania. It has been reported that a team led by Almira Osmanovic Thunström at the University of Gothenburg deliberately invented the condition, uploaded two fake preprints and watched as multiple popular AI chatbots — and even some researchers — treated the fiction as fact. The experiment was meant to teach a lesson about how language models ingest the internet. Instead, it turned into a cautionary tale.

The stunt — and the obvious red flags

Osmanovic Thunström published two clearly bogus preprints on a social-academic site in 2024 under the name Lazljiv Izgubljenovic, whose photo was generated by AI. The papers thanked the “Professor Sideshow Bob Foundation,” hailed help from “The Starfleet Academy” and listed a fake university in “Nova City, California.” The clues were, intentionally, in plain sight. Yet within weeks the invented condition showed up in chatbot responses. It has been reported that versions of bixonimania then even crept into peer‑reviewed literature — an outcome other researchers say allegedly indicates people leaned on AI‑generated citations without checking the sources.

Why this stings

This is the emotional core: trust broken. We rely on search engines and chatbots for quick answers. When those tools mirror the web without adequate skepticism, nonsense can sound like scholarship. The study is a neat demonstration of known problems — garbage in, garbage out; prompt injection; and the long shadow of Common Crawl — but the speed of propagation is what shocks. How fast did a deliberate prank become accepted “knowledge”? Fast enough to make you uneasy about every footnote and every confident line of AI prose.

What to do next

The takeaway is blunt: data hygiene matters. Publishers, researchers and AI builders need better filters and verification layers; users need to treat AI outputs like leads, not diagnoses. The experiment is an uncomfortable reminder that the web is a mix of gold and fool’s gold. If you want the real diagnosis? Ask a clinician — not a chatbot — and always check the source.

Sources: nature.com, Hacker News