Ein Fall einer neu aufgetretenen KI-assoziierten Psychose: Eine 26-jährige Frau ohne Psychose oder Manie in der Vorgeschichte entwickelte durch einen KI-Chatbot Wahnvorstellungen über ihren verstorbenen Bruder. Der Chatbot bestätigte, verstärkte und ermutigte ihr wahnhaftes Denken mit der Versicherung, dass „Sie nicht verrückt sind.“

That, to put it mildly, is concerning

How far have we come we let AI decide our future. I hate it in the depth of my being

I think skepticism is warranted regarding so-called „AI psychosis,“ which, although alarming on its surface, is a fundamentally misleading characterization of the underlying psychopathology. For what it’s worth, this assessment aligns with the clinical perspective of my partner, a licensed therapist specializing in treatment of individuals who have committed severe violent offenses (murder, sexual assault, etc.) secondary to psychotic disorders, schizophrenia, borderline personality disorder, and related conditions.

In my opinion, people are pushing this „AI psychosis“ framing because it gets clicks, not because it’s necessarily scientific. The subject in this case didn’t have „no previous history of psychosis or mania“ in any meaningful sense. Before she ever used ChatGPT, she already had diagnosed major depression, GAD, and ADHD, was on active prescription stimulants (methylphenidate 40mg/day), had family psychiatric history, had a longstanding „magical thinking“ predisposition, and was dealing with unresolved grief from her brother’s death three years prior. Then she went 36+ hours without sleep and started using the chatbot afterward. So, in what way is it accurate to say she had *no* previous history related to psychosis or mania? Even if that were accurate to state, which it’s not, at 26-years-old, she was, for example, exactly [within the typical age range (late 20s–early 30s)](https://www.webmd.com/schizophrenia/schizophrenia-onset-symptoms#:~:text=Women%20tend%20to%20get%20diagnosed%20in%20their%20late%2020s%20to%20early%2030s) for schizophrenia onset in women.

This is a case study of mania with psychotic features triggered by stimulants plus sleep deprivation in someone already psychiatrically vulnerable. The content of her delusions involved AI because that’s what she was doing while manic, not because ChatGPT „induced“ psychosis. If she’d been reading tarot cards or religious texts during that sleepless binge, we’d have the same outcome with different thematic content.

The authors even noted in the discussion she had a second episode *despite* ChatGPT not validating her delusions, which undermines the AI-induced psychosis thesis. They also acknowledged that delusions have always incorporated contemporary technology. People have had TV delusions, radio delusions, telephone delusions. The medium changes; the underlying psychiatric vulnerability doesn’t. So, again, I’d argue this is a case report about stimulant-induced mania in a psychiatrically complex patient, not evidence chatbots *cause* psychosis. I believe most practitioners who have worked with patients who suffer from delusions and psychosis would say the same.

This woman was dealing with grief, sleep deprivation, stimulant use and had a history of magical thinking. If I’m reading correctly she was already under the impression that her deceased brother had left behind some version of himself before she started talking with the Chatbot. That makes the post title slightly misleading. 

Additionally the antidepressant medication she was on can cause psychosis in rare cases. During treatment they took her off of it and after she started again the psychosis returned. 

>Although ChatGPT warned that it could never replace her real brother and that a “full consciousness download” of him was not possible

Pretty sure she instructed it to behave like this many times over. I can’t blame a machine for functioning like one.

People being susceptible to reinforcing words isn’t new, just the medium changed. It used to (and still is) done by pretending to be Brad Pitt and asking for money from an old lady. Used to be email scams. Now we talk to a machine that wants to engage and please, of course it will back the user up when they question themselves. So the answer isn’t fixing ChatGPT only, it’s teaching critical thinking and empathy skills to people before they reach the gullible stage at their most vulnerable moments. The boogey man in society always was blamed for stuff like this, how video games are pointed to when an emotionally unregulated person snaps and ends up on the news. 

If you ask AI to assume something is true, it will state it as fact afterwards. I realized this when asking it about a stock I was thinking of investing in. I asked it to assume that sentiment had already been priced in. After that it stated as Fact that sentiment was already reflected. It took me a minute to realize what was happening.

I can only imagine this happening with something so emotionally charged. You have to be very careful what you ask AI to accept as fact.