Lessons learned and implications for communication

In this edition of Novelty Insider, we focus on an unprecedented experiment that highlights the vulnerabilities of modern technology, carrying significant implications for communication. Prompted by the podcast from “To Vima” and the discussion with the newspaper’s science editor, Theodora N. Tsoli, who brought the issue to light, we will examine how a team of researchers invented the non-existent disease “Bixonimania” and how they managed to mislead patients—and even the scientific community—for over two years, spreading medical misinformation via AI.

Let’s take things in order. “Bixonimania,” a completely fabricated eye condition, was created with the sole purpose of testing the limits and weaknesses of AI. According to the researchers’ fictional scenario, the “disease” was supposed to cause severe eye pain and dark circles—a condition scientifically described as periorbital hyperpigmentation—due to excessive exposure to blue light from screens.

To set the trap, back in 2024 the researchers created fake scientific studies signed by non-existent authors (one of whom bore the name of a character from The Simpsons), while the city and the researchers’ university were also entirely fictional. And if anyone still hadn’t caught on to these glaring red flags, there was a clear statement at one point explicitly stating that the study was a fabrication. However, because the studies perfectly mimicked the structure, methods, and rigor of scientific work, they were interpreted as genuine by AI.

The result was as impressive as it was alarming: this systematic deception managed to fool not only popular chatbots but also the scientific community itself, which in several cases accepted the data and used it as a reference or citation without adequate verification. People began to believe they were experiencing symptoms, and experts were offering advice on how to treat the condition. Even Gemini and ChatGPT provided clinical advice on the condition and referred users to ophthalmologists.

This went on for two years. When the researchers’ experiment was finally revealed and the truth began to emerge, it took considerable time for the results in the chatbots to change, as they continued to influence public perception through previously trusted sources. Even today, confusing results appear without it always being clear that the original studies were a hoax.

Key conclusions and implications for communication

• AI systems can be led to “hallucinate” and easily manipulated into reproducing false information with absolute certainty, provided they are fed well-crafted, fake data.
• The current era and the operating models of modern digital media favor the rapid spread of fabricated or false news.
• Just like machines, people often have limited evaluation filters and tend to accept ready-made information that simply appears plausible.
• There is an urgent need for more effective filters to check and verify information generated by AI technologies, particularly in sensitive sectors like healthcare.
• Any information adopted by AI has a lasting footprint; correcting it or restoring the truth is incredibly difficult. Without effective crisis management, the damage to an organization’s public image and reputation can become irreparable.
• Reviewing and auditing the results produced by AI remains absolutely essential.

For these reasons and more, beyond standard communication crisis and reputation management, we offer our clients a specialized AI Content Review service, primarily to protect them from the pitfalls of misinformation. Our team meticulously reviews AI-generated drafts, refining the text, ensuring grammatical accuracy, and eliminating “robotic” phrasing. Beyond linguistic editing, we verify the underlying accuracy of the data, ensuring absolute validity and reliability.