The bixonimania experiment was not designed to embarrass a single chatbot — it was designed to test whether the AI-assisted information layer has any mechanism for forgetting a false entity once it has been absorbed. The team uploaded fabricated studies to public repositories, waited for the indexing cycle, then queried multiple popular chatbots about the invented condition. The models responded with descriptions, symptom profiles, and management suggestions for a disease that Osmanovic Thunström's team had conjured from scratch. When the studies were later removed, the condition remained in model outputs.…