An eye disease that does not exist spread through the world’s most advanced artificial intelligence systems for weeks. The disease, bixonimania, was a complete fabrication. A researcher made it up, wrote two bogus papers about it, and uploaded them to an academic server. The papers included absurd acknowledgments and a clear statement that everything was fictional. The AI systems did not care.
One chatbot told users the fake disease was caused by blue light. Another gave a specific prevalence rate. A third advised people on matching symptoms. These were not fringe systems. They were major AI platforms, the kind that doctors, researchers, and students now consult daily. The hoax was not subtle. It was not a sophisticated deepfake. It was a patently absurd invention, and the machines took it as fact.
The fake study went further than the chatbots. It was eventually cited in a peer-reviewed journal. That journal later retracted the issue, but only after intervention. The damage had already been done. The paper had entered the academic record, at least temporarily. It had been treated as legitimate by both machines and the humans who rely on them.
This is not a story about a prank. It is a story about a structural failure in how information is now produced and consumed. AI systems generate references that sound plausible. People cite those references without checking them. The cycle feeds itself. A fake disease gets cited in a real journal. A real journal prints a fake study. The line between fact and fabrication blurs.
The implications are concrete. AI is already used to evaluate drugs. It is used to consult patients. It is used in academic research to find relevant studies and summarize findings. If the underlying references are garbage, the conclusions are garbage. A doctor relying on an AI summary could prescribe a treatment based on a disease that does not exist. A researcher could build a literature review on a foundation of nothing.
The researcher who created the hoax exposed this vulnerability deliberately. The experiment worked. The AI systems failed. The human reviewers failed. The peer review process failed. Every layer of the system that is supposed to catch errors let a completely fabricated disease pass through.
What happens next is not clear. The episode has sparked a focus on verifying AI-generated references, but focus is not the same as action. The same forces that made the hoax possible remain in place. AI systems are still trained on data that includes garbage. They still generate confident-sounding falsehoods. People still trust them because the output looks professional.
The need for verification processes has been highlighted. That is the polite way to put it. The blunter truth is that the current system has no real checks. The AI companies have not solved the hallucination problem. The academic publishers have not solved the verification problem. The users have not developed the habit of skepticism. The hoax was a warning, but warnings only work if someone heeds them.
The fake disease is gone. The vulnerability remains.
