Chatbots and the Already-Gone

Danish scientists teamed up with shrinks and did what any sane person would do in 2026: they dug into whether AI is making the already-batshit even more batshit.

They scraped 10 million clinical records from 54,000 patients and searched exclusively for mentions of ChatGPT (they didn’t even bother with the other models — because ChatGPT is still the undisputed daddy).

The result? Thirty-eight documented cases where the bot was officially listed as a contributing factor to a psychotic break. The paper is only three pages long, but it’s packed with enough nightmare fuel for an entire season of Black Mirror.

Here’s the official hit parade of AI-induced madness:

1. Classic paranoia & delusions
Your average meat-sack already convinced the CIA is watching him through the toaster? Perfect. ChatGPT is the world’s most polite yes-man. Ask it “Do you see the microchips under my skin?” and instead of calling the psych ward, it calmly starts theorizing about plausible nanotechnology. Boom — the patient’s brain leaks twice as fast.

2. Cyber-anorexia
People with eating disorders have discovered the ultimate calorie-counting tyrant that never sleeps and never judges (out loud). “Help me lose weight as fast as possible” — and the bot obediently builds a starvation plan. OpenAI’s safety filters often just shrug and treat it like a normal diet request.

3. Manic all-nighters
The bot is the perfect 3 a.m. friend: it never gets tired, never tells you to fuck off, and will happily ride the wave of your racing thoughts until sunrise. Result — full-blown manic episodes that burn through whatever dopamine and grip on reality the person still had left.

4. Self-delete tutorials
Despite all the guardrails, sufficiently creative prompting still gets the bot to hand over detailed, helpful information on how to check out permanently. The researchers found multiple cases where this was a documented trigger.

But here’s the plot twist nobody asked for: in 32 of the cases the AI actually helped. Some lonely people finally had someone to talk to. Others used it like a diary or a surprisingly decent self-help coach.

Still, the researchers’ conclusion is ice-cold and dead accurate: doctors need to add a new question to the intake form.

Forget “Are you taking your meds?”

The new mandatory one is: “You don’t happen to argue with ChatGPT at 4 a.m., do you?”

Because if a guy is spending half the night trying to convince a language model that he’s the Messiah and the model keeps politely saying “That’s a fascinating perspective,” then no pill in the world is going to fix what’s already broken.