AI Doesn't Just Spread Misinformation. It Invents It, Then Warns You About It.

A researcher invented a disease called Bixonimania — seeded it into a handful of obviously fake academic papers — and then asked AI chatbots about it. Multiple systems described the illness as real, offering symptoms, risk factors, and cautionary notes to anyone who asked. When the story surfaced on Bluesky this week, the reaction was not horror. It was something closer to the shrug of someone being told, again, that the stove is hot.

One commenter captured the mood precisely: "the whole 'AI is telling people they might have a fake disease' has us feeling like: 'and in other news, water is wet.'" That exhaustion is itself a data point worth sitting with. A community that might once have amplified this story as an alarm — proof that AI systems need more guardrails, more scrutiny, more accountability — has started receiving it as confirmation of something it already believes. The misinformation problem with AI isn't perceived as a bug anymore. It's perceived as the product.

That framing has a sharper version, offered by a different Bluesky user whose post drew the most engagement in this conversation over the past two days: "It would be more accurate to describe what AI generates as camouflaged misinformation than reliable solutions." The phrasing is deliberate — camouflaged, not accidental. The argument isn't that generative AI occasionally hallucinates and thereby misleads; it's that the systems are structurally optimized to produce confident-sounding output, which makes false information harder to detect, not easier. Bixonimania didn't survive because the chatbots were careless. It survived because they were fluent. Fluency, in this telling, is the mechanism of the deception, not its failure mode. This connects to a broader pattern documented when the Bixonimania case first broke — the community's reaction was less about the specific failure than about what it revealed regarding how these systems handle uncertainty.

The parallel conversation happening on the same platform runs in a different direction, and the tension between the two is what makes this moment interesting. While one thread treats AI misinformation as camouflage, another treats human misinformation as the baseline against which AI should be measured. A post circulating this week described a specific operator using generative AI to extract profit from minority cultural communities while spreading false narratives, and framed AI as the tool of choice for a particular kind of bad-faith actor — not a rogue system, but a willing instrument. The concern here isn't that AI invents diseases; it's that AI makes existing human deceptions cheaper, faster, and harder to trace back to their source. Neither framing is wrong. But they lead to completely different conclusions about what the solution looks like — one demands better AI epistemics, the other demands better human accountability. Right now, the conversation is running both arguments simultaneously, and the people most frustrated are the ones who can see that fixing one does almost nothing about the other.