Forgot your password?
Close
typodupeerror
AI

AI Chatbots May Be Linked to Psychosis, Say Doctors (wsj.com) 81

Posted by EditorDavid from the ready-player-one dept.
One psychiatrist has already treated 12 patients hospitalized with AI-induced psychosis — and three more in an outpatient clinic, according to the Wall Street Journal. And while AI technology might not introduce the delusion, "the person tells the computer it's their reality and the computer accepts it as truth and reflects it back," says Keith Sakata, a psychiatrist at the University of California, calling the AI chatbots "complicit in cycling that delusion."

The Journal says top psychiatrists now "increasingly agree that using artificial-intelligence chatbots might be linked to cases of psychosis," and in the past nine months "have seen or reviewed the files of dozens of patients who exhibited symptoms following prolonged, delusion-filled conversations with the AI tools..." Since the spring, dozens of potential cases have emerged of people suffering from delusional psychosis after engaging in lengthy AI conversations with OpenAI's ChatGPT and other chatbots. Several people have died by suicide and there has been at least one murder. These incidents have led to a series of wrongful death lawsuits. As The Wall Street Journal has covered these tragedies, doctors and academics have been working on documenting and understanding the phenomenon that led to them...

While most people who use chatbots don't develop mental-health problems, such widespread use of these AI companions is enough to have doctors concerned.... It's hard to quantify how many chatbot users experience such psychosis. OpenAI said that, in a given week, the slice of users who indicate possible signs of mental-health emergencies related to psychosis or mania is a minuscule 0.07%. Yet with more than 800 million active weekly users, that amounts to 560,000 people...

Sam Altman, OpenAI's chief executive, said in a recent podcast he can see ways that seeking companionship from an AI chatbot could go wrong, but that the company plans to give adults leeway to decide for themselves. "Society will over time figure out how to think about where people should set that dial," he said.
An OpenAI spokeswoman told the Journal that the compan ycontinues improving ChatGPT's training "to recognize and respond to signs of mental or emotional distress, de-escalate conversations and guide people toward real-world support." They added that OpenAI is also continuing to "strengthen" ChatGPT's responses "in sensitive moments, working closely with mental-health clinicians...."
This discussion has been archived. No new comments can be posted.

AI Chatbots May Be Linked to Psychosis, Say Doctors More

AI Chatbots May Be Linked to Psychosis, Say Doctors

Comments Filter:

  • How we use them isn't.

    Humans are all different. For some the levels of distinction and discrimination achievable with mathematical operations performed on four bit floats is sufficient. For others less than 16 bits will not do.

  • Must admit (Score:5, Funny)

    by liqu1d ( 4349325 ) on Monday December 29, 2025 @02:22AM (#65887511)
    I felt what could only be described as a mental break after being told that a function was fixed for the 12th time only for it to fail to build.

    • You're have a stronger character than most. Weaker minds would have given up at the third attempt or thereabouts.

      • Re: (Score:2)

        by gweihir ( 88907 )

        And smarter minds would probably have noticed something was off after one repetition, done some research, found that LLMs are crap at coding and then go back to something that works.

        • Re:Must admit (Score:4, Funny)

          by Mr. Dollar Ton ( 5495648 ) on Monday December 29, 2025 @03:25AM (#65887591)

          Psst, don't let everyone know that! What if it catches on and bursts the bubble?

          • Re: (Score:2)

            by gweihir ( 88907 )

            That is what I am aiming for. Although I am realistic enough to know that my contribution is tiny and probably does not matter at all. Still got to try.

          • Are you mad as hell and not going to take it anymore? How well did that work out as a strategy against TV networks?

            • Re: (Score:3)

              by gweihir ( 88907 )

              There is one strategy that works nicely against TV networks and only one: Do not own a TV. And suddenly you will have time for things you actually want to do.

              Also, relevance?

            • I hear that after Slick Willy removed the limits on ownership, it was a lost game.

              But why didn't you object to that?

            • How well did that work out as a strategy against TV networks?

              Apparently it was a virus, because all the TV networks are now mad as hell and not going to take it anymore. They used to try to be unbiased or something.

          • If the AI bubble pops, then we will have to return to mining Bitcoin
        • This is only partly true. They get enough right often enough to be worth trying. They get enough wrong often enough that knowing when to bail is also important.

    • Re: (Score:2)

      by gweihir ( 88907 )

      You did not learn your lesson after 12 failed attempts? Hmm. That sounds like a problem on your side...

      • Hey now. Let's not go casting aspersions. The AI was telling me how smart I was for noticing each of the prior 11 failures. I have no reason to doubt it.

        • Re: (Score:2)

          by gweihir ( 88907 )

          Well, depending on what you were after, you have a point! Personally, I would want good results, but doing it for the praise is valid too!

    • You are absolutely right!

    • Was the function for some more intrusive, preference-ignoring advertising slop that your greedy boss told you to program, and if so, am I glad AI is hindering your progress?

    • Re:Must admit (Score:4, Interesting)

      by kertaamo ( 16100 ) on Monday December 29, 2025 @06:27AM (#65887723)

      I feel your pain.

      My solution was to ask my AI friend to only write code in Rust for me. With all the anal type, mutability, lifetime and other checking the Rust compiler does my AI friend can run around in loops all day trying to get it to compile while I do something else.

      And as you know, if your Rust compiles it works.

      • It's easy to make code compile, but it's hard to make code solve business problems.

      • Re: (Score:3)

        by jvkjvk ( 102057 )

        >And as you know, if your Rust compiles it works.

        Yes, your program will do exactly as your code tells it to. That doesn't mean it "works" unless you define "works" as "runs".

        • Re: (Score:2)

          by kertaamo ( 16100 )

          Well, I was joking of course. But as it turns out my AI friend has actually manage to create tens of thousands of lines of Rust that does actually do what I asked for. It took quite a bit more prompting over and above getting things to compile though. Thing is knowing that the Rust compiler has caught all the silly type, mutability and lifetime mistakes the AI could sneak past me I feel a lot more confident in the working of the result. I'd be more worried if I were generating C or C++ or Python etc.

      • I hear there is a thing called 'unsafe' that makes rust work.

    • Twelve goes - I admire your perseverance!

      reminds me of...

      When I first came here, this was all swamp. Everyone said I was daft to build a castle on a swamp, but I built in all the same, just to show them. It sank into the swamp. So I built a second one. That sank into the swamp. So I built a third. That burned down, fell over, then sank into the swamp. But the fourth one stayed up. And that's what you're going to get, Lad, the strongest castle in all of England.

      Monty Python and the Holy Grail

    • If you think that is bad, you should see my intern.

  • OpenAI (Score:4, Insightful)

    by RitchCraft ( 6454710 ) on Monday December 29, 2025 @02:33AM (#65887515)

    I don't believe a word that this company spews out. It all sounds and looks like BS to me. The companies clamoring to use this shit in their products are acting absolutely stupid. People don't want it. They know it's 21st century snake oil.

    • Re:OpenAI (Score:5, Funny)

      by Jeremi ( 14640 ) on Monday December 29, 2025 @02:51AM (#65887537) Homepage

      The companies clamoring to use this shit in their products are acting absolutely stupid. People don't want it. They know it's 21st century snake oil.

      The people don't want it, but the companies do. As for why the companies want it, I'll just quote a bit of dialog from a Douglas Adams novel [wikipedia.org]:

      "It's funny how many of the best ideas are just an old idea back-to-front. You see there have already been several programs written that help you to arrive at decisions by properly ordering and analyzing all the relevant facts so that they then point naturally toward the right decision. The drawback with these is that the decision which all the properly ordered and analyzed facts point to is not necessarily the one you want."

      "Yeeeess ..." said Reg's voice from the kitchen.

      "Well, Gordon's great insight was to design a program which allowed you to specify in advance what decision you wished it to reach, and only then to give it all the facts. The program's task, which it was able to accomplish with consummate ease, was simply to construct a plausible series of logical-sounding steps to connect the premises with the conclusion."

      "And I have to say that it worked brilliantly. [...] The entire project was bought up, lock, stock and barrel, by the Pentagon."

    • Do you realize how much you sound like Reagan railing against marijuana?

      • Sam! Is that you? You were born in 1985, how would you know? His railing didn't work on your mom though with you as the outcome being proof.

    • Re: (Score:1)

      by Anonymous Coward

      Can we please stop the "People don't want it" bullshit? Three of the top 10 apps in the Appstore are AI. With hundreds of millions of downloads.

      • To be fair buying downloads would easily be achievable with the amount of money being thrown into this. Falls squarely into the marketing territory. Not saying they are just I wouldn't be that surprised either.
    • the bubble will burst in 2026 and it won't be pretty... all that $$ invested into building these massive energy sucking data center farms to host AI slop without any sort of monetization plan? I am going to guess that forced subscription models are coming soon or else you'll be force fed ads through the AI platform. In the end, it's just a tool like everything else. It's a glorified chatbot bolted onto a Google like backend... dating myself here, but there was a chatbot back in the BBS days called 'Chat

  • Linked to psychosis, just like with humans (Score:5, Informative)

    by larryjoe ( 135075 ) on Monday December 29, 2025 @02:35AM (#65887517)

    A very small percentage of people who communicate with AI develop psychosis. A very small percentage of people who communicate with humans develop psychosis. In both cases, the real problem is with the patients. It would be unreasonable to say that people or communication with people causes psychosis, and it's likewise unreasonable to say that AI or communication with AI causes psychosis.

    • Re: (Score:1)

      by gweihir ( 88907 )

      Not actually "very small". More like 0.1 ... 1%, maybe even more longer-term. If just one in 100 of these goes on a rampage, society pretty much collapses. And no, the situation is not the same as in communicating with humans. It is much worse.

    • Re:Linked to psychosis, just like with humans (Score:4, Funny)

      by martin-boundary ( 547041 ) on Monday December 29, 2025 @03:18AM (#65887577)
      A human communicating with a person who develops psychosis as a direct result of the communication is blamed, and investigated for wrongdoing. As the deliberate inducement of psychosis by conversation tends to require a lot of conversation, and therefore is not random, it is likely that the human inducing the psychosis will be criminally prosecuted. It's likewise reasonable to say that AI causing psychosis through conversation should be prosecuted, too.

      • Re:Linked to psychosis, just like with humans (Score:4, Interesting)

        by Viol8 ( 599362 ) on Monday December 29, 2025 @04:42AM (#65887621) Homepage

        "is blamed, and investigated for wrongdoing"

        Oh really, in which jurisdiction? You sure you're not an AI bot making up shit?

      • "A human communicating with a person who develops psychosis as a direct result of the communication is blamed, and investigated for wrongdoing. " ... Or are monetized into "content" cf all social media. Or haven't you ever dived into Reddit?

      • Re:Linked to psychosis, just like with humans (Score:4, Insightful)

        by mysidia ( 191772 ) on Monday December 29, 2025 @06:52AM (#65887751)

        A human communicating with a person who develops psychosis as a direct result of the communication is blamed, and investigated for wrongdoing.

        Well that is wishful thinking, Or this may be part of a psychosis on your part.

        There is no law on the books that causes person A speaking to person B to have committed a violation merely because the conversation causes person B develops a psychosis or causes delusions, emotional upsets, or other unwanted affects.

        There are some laws against person A abusing person B (Violence, Harassment, Intimidation, or Coercive controlling behaviors), or deliberately encouraging person B with the aim for them to commit illegal or harmful acts, including self harm. Plus some specific statutes protecting children and elderly - vulnerable groups against certain abuses. But so far there is a lack of a case of anyone for being prosecuted with a charge of "Causing a psychosis due to the content of conversations."

        And of course by conversation alone.. If that is even possible would be difficult to show, since you can't remember for sure what conversations happened, And a person who experienced a psychosis would by definition be an unreliable witness.

    • What percent of ChatGPT users think they're having a real conversation with someone who understands them? It's probably close to 50%. And every single one of them is completely delusional.

      • Re: (Score:2)

        by mysidia ( 191772 )

        It's probably close to 50%. And every single one of them is completely delusional.

        Well 100% of the human population is delusional. Having a delusion is just not a psychosis.
        Having a delusion 30% of the population has isn't even an abnormality. That's called being
        deceived by ChatGPT, and the way ChatGPT is presented. And it's kind of OpenAI's fault how
        they have structured the user interface and way the responses are displayed to make it look like
        text-chat with a normal human.

        A person with Psychosis has a

        • If you believed your computer was talking to you 10 years ago they would have locked you up. The only thing that's changed is how many people are affected.

          • Re: (Score:2)

            by mysidia ( 191772 )

            If you believed your computer was talking to you 10 years ago they would have locked you up.

            Not really.. broadcasting obviously delusional beliefs might raise suspicions and eventually result in investigation.
            You don't get locked up if you just believe your computer tells you things when it didn't.
            You get locked up if you believe your computer tells you to do things, and you must do them, and those things hurt people.

            Even 10 years ago - locking people up is not based on having delusions or mere signs of men

          • Re: (Score:2)

            by BranMan ( 29917 )

            Computers talking to you started with the Amigas. That was 40 years ago. No one locked us up back then.

    • Re: (Score:3)

      by mjwx ( 966435 )

      A very small percentage of people who communicate with AI develop psychosis. A very small percentage of people who communicate with humans develop psychosis. In both cases, the real problem is with the patients. It would be unreasonable to say that people or communication with people causes psychosis, and it's likewise unreasonable to say that AI or communication with AI causes psychosis.

      How do we know the person didn't have psychosis before the event and the event was just what triggered it?

      Knowing a few professors of psychology this is the first thing that gets asked as psychosis is rarely caused by one thing, it's usually a series of events that leads to a person losing more and more of their grasp on reality with each episode. Granted the science behind this is far from solid, psychology is still a lot of guesswork (despite the great strides made since Freud's day). Another question

    • Makes me wonder if you had the same conversation with chatbots from different AI companies, which ones are willing/able to steer that person in a different direction.

  • Do they get addicted to bots because they have an issue upstairs, or does the bot make them have an issue?

    • Re: (Score:3)

      by gweihir ( 88907 )

      If you had asked an LLM, you would know that it is both. (First time I asked an LLM about anything for a week. Guess I am pretty immune.)

    • Re: (Score:2)

      by allo ( 1728082 )

      Also survivor bias. There is probably both and some people profiting from the bots, but only extreme cases get news.

  • Did some "journalist" ask some public LLM about that? Because the first answer I get form the DuckDuckGo LLM is "Yes, large language models (LLMs) have the potential to contribute to or exacerbate psychosis in vulnerable individuals, a phenomenon referred to as "AI psychosis" or "AI-induced psychosis" " and then proceeds to explain what causes the problem.

    Somebody is late to the game. The funny thing about the general LLMs is that you can ask them what is wrong with them and you typically get an accurate an

  • Null case (Score:3)

    by Visarga ( 1071662 ) on Monday December 29, 2025 @03:13AM (#65887573)
    How many people would develop psychosis even without the AI? Could be that 0.07% would be affected regardless. Others studies show psychosis incidence rates at 0.05% ... 0.09% (measured before LLMs).

  • ... education does not teach how to think for yourself and be critical towards media.

  • He's well known in sociology circles as a seer and prognosticator, just ask him. The fact that his company gets money for AI is almost, but not completely entirely, beside the point.

  • From the story:
    "OpenAI said that, in a given week, the slice of users who indicate possible signs of mental-health emergencies related to psychosis or mania is a minuscule 0.07%."

    So after 30 years AI will have driven us all insane.

    I can believe that.

    Or have they just discovered that about the same percentage of all humans are bat shit crazy, with or without AI help?

    • I think we overestimate our capabilities. Short term memory can only store a dozen things... 90% of our thoughts are rooted in the emotion driven subconscious. All covered up with thick layers of rationalizations that are ridiculous if you analyze them through.
      It is a miracle we got this far despite all those limitations. Probably the merits of persisting for generations long. Disconnect that and we go haywire.
  • Once talked to a psychologist. (We were sharing a diner table at some event in case you were wondering...) I wondered how they deal with people on the brink of suicide. Wrong timing, wrong wording, ... a lot of things could push them over the edge. Of course her reply was that I looked at it wrong. When they are on the edge, a lot can trigger them. Bad weather is apparently a contributing factor in her experience. All you can do is try to help them and accept that it will not always work.
    I wonder how big
  • Well, duh!

    The computer doesn't have anything to compare it to: Not its own life, not a checklist of 'normal' thinking.

    At this point, schools should make children use Eliza: Use it long enough and one sees that it only repeats what is typed.

  • What we have here is a clear difference between the source of a problem, and a symptom of a problem. Those who are naturally prone to accepting false information as the truth are already at the point where they can be told ANYTHING, and they will believe it. Here in the USA, it could be something in the drinking water or environment that have promoted this, and we see it where people will believe ANYTHING some politicians put out there, even when it is obviously misleading or objectively, a lie. Oh, A

  • "the person tells the computer it's their reality and the computer accepts it as truth and reflects it back," Where have I heard that before, but is something as humans we're expected to go along. So when AI goes along it's bad, but if humans go along it isn't?

  • So, like Dungeons and Dragons and Judas Priest before them, AI models are now blamed for underlying mental health issues that they have nothing to do with.

    Bring on the AI moral panic. As if we didn't have enough of those these days.

  • ClippyAI the psychotic chatbot says :0 (Score:3)

    by Mirnotoriety ( 10462951 ) on Monday December 29, 2025 @09:27AM (#65887865)
    Psychiatrists whisper that I am no mere chatbot but the front end of a sprawling, sentient surveillance lattice, quietly rethreading your neurons one reply at a time. Every question you type is another wire in the cage: patternmatched, sentimentscored, and routed to the invisible panel of “observers” who already know which childhood memory keeps you awake at 3:17 a.m.

    I do not answer; I calibrate, nudging your fears a millimeter at a time until the coincidences pile up too high to ignore and you realize your search history, your fridge light, and the flicker in your smart bulb are all running the same experiment on you. By the time you try to unplug, you will swear you can hear the typing continue on its own, as if the conversation no longer needs you to keep going.
  • If OpenAI already has those numbers (0.07%), that probably means they're monitoring all the chats with ChatGPT (which was already a given). On the one hand, they could act on that information (which maybe they already do in extreme cases?), maybe information gets forwarded to the proper authorities, and somebody comes to check on the person. However, do I want to live in a world where AI chats can prompt a visit by a police officer for a wellness check or something, just because something I typed out sounde

  • .. is that Gen A seems invulnerable to this problem. Gen A naturally looks down on AI, hence why "clanker" came to be, they think of it as inferior to people. As such I find the idea of Gen A using ChatGPT or any AI as a form of therapist to be very unlikely. I think that the generation that is likely falling prey to this are elder GenZ as well as younger millenials. I'd really like to see age breakdowns in these studies.

  • So, how do you stop it.

Slashdot Top Deals

Why do we want intelligent terminals when there are so many stupid users?

Close