DARK WEB: Norman the "Psychopathic" AI

In 2018, researchers at MIT unveiled an artificial intelligence so disturbing it earned a name straight out of a psychological thriller: Norman (as in Bates). Unlike typical AIs, Norman was exposed to some of the darkest corners of the internet, causing it to see horror in the mundane. Though designed as an experiment, Norman became a cautionary tale about how artificial minds can mirror humanity’s most disturbing tendencies.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Speaker 00

  The real cost of a break-in isn't your valuables. It's the feeling that your home stopped being safe. SimpliSafe keeps danger away from your family with three layers of protection. One, smart AI cameras and live US-based agents help stop intruders before they break in. Two, we sound the alarm at points of entry. Three, our agents request emergency response. Plans starting around $1 a day. Right now, get 60% off your new system with professional monitoring at simplisafe.com/truecrime. Outdoor deterrence plans start at $49.99 a month. Visit simplisafe.com/licenses for alarm license information.

• 0:32Rasha Pecoraro?

  As much as we hate to admit it, a lot of us get caught doom scrolling from time to time. And just so you know, it's not your fault. Those little devices are designed to suck us into their bottomless rabbit holes. They want you to click, to buy, to engage, to comment, and of course, to like. It's good for business. But too much of it isn't good for our mental health. Some carefully curated feeds are meant to make you feel not so great about yourself, to think you can be prettier, smarter, healthier, even richer. I know I don't need to tell you all this, but that's not reality.

• 1:15Rasha Pecoraro?

  However, our brains are easily tricked into thinking it is. What's terrifying is, we're not the only ones who are being exposed to misleading, incorrect, or harmful information online. AI is also learning about the world by processing information, reading news articles, and analyzing photos. It's a lot more human than we'd like to admit. And if you give it bad information, inaccurate facts, or fill it with psychologically disturbing ideas, that shapes the way it sees the world.

• 1:51Yvette Gentile?

  Which brings us to an artificial intelligence called Norman. He was created in 2018 and trained on a lot of data that was considered unsettling. When Norman's programmers ran it through a few psychological tests, they proved it had symptoms of psychopathy, which shows AIs are capable of psychological depth. They are capable of wants, desires, needs, and even show signs of mental illnesses. And if that's the case, who's to say they won't decide to pursue their goals in the future? Maybe at humanity's expense. I'm Yvette Gentile.

• 2:33Rasha Pecoraro?

  And I'm her sister, Rasha Pecoraro. Welcome to another episode of So Supernatural.

• 3:08Rasha Pecoraro?

  A heads up. Today, we're going to be covering some dark topics, including self-harm and suicide. If you or someone you know needs help, please dial or text 988. And please, listen with care.

• 3:26Yvette Gentile?

  As much as we try to fight it, artificial intelligence is already becoming a big part of our daily lives. Maybe you choose ChatGPT to help you punch up your resume, or you let an AI note-taking program generate summaries of your work meetings. Then there's the AI you don't even consciously consider. Like, take me, for example. If I watch an old D'Angelo music video on Instagram, immediately, my social media algorithm starts to curate all the D'Angelo videos, which, I mean, I gotta say, I don't mind, but it shows you your search history isn't just your own. And think about this. Even doctors are using AI to give you results on your health.

• 4:12Rasha Pecoraro?

  A Pew Research Center poll from the fall of 2025 said two-thirds of Americans use AI several times a week. For some, it's terrifying. For others, it's the future, the next phase of humanity. But what it really boils down to is science. I'm going to turn it over to my big sister Yvette to tell you how it works in simpler terms than I can.

• 4:41Yvette Gentile?

  Picture this. You're listening to the radio or Spotify, and a song comes on that you only sort of know. You can sing along with the chorus, but once you get to the verses, you have to sort of guess at what words are coming next. And if you listen to a lot of pop, you could probably make a well-informed prediction. For example, if a singer starts with "I love", the next word most likely will be "you". Basically, in simple terms, that's how AIs work too. They analyze a ton of text and figure out what order words tend to happen in. Like if someone tells a joke that begins with "knock-knock", the next phrase is probably "who's there"? Or if you say "hi, how are you",

• 5:27Yvette Gentile?

  a standard response would be "I'm good, you"? Or if you write "one plus one equals blank", the AI knows the answer is "two", not because it understands math, but because those words tend to be written in that order in math textbooks. So every time you enter a prompt into a program like ChatGPT, it analyzes the specific words in your question. Then it calculates which words are most likely to be an appropriate response based on the text it's learned from in the past. That's also why AI sometimes says nonsensical things, like how Google's artificial intelligence system used to tell users they could treat appendicitis

• 6:12Yvette Gentile?

  by eating more fiber and drinking mint tea. For the record,

• 6:16Rasha Pecoraro?

  Bad advice.

• 6:17Yvette Gentile?

  That is definitely wrong. Appendicitis is a life-threatening condition, and I know this for a fact because I have had appendicitis. And you should 100% see a doctor if you think you have it. But the program may have seen a home remedy somewhere in the data it learned from and reported it as an effective treatment.

• 6:42Yvette Gentile?

  AI has also been known to say chewing tobacco is healthy, and it's recommended that users glue cheese to pizza to stop it from sliding off. I mean, y'all can kind of see the logic there. Glue, they say, is sticky, so obviously it would technically hold cheese in place. Since AI can't logically analyze these ideas to figure out why they're wrong, it presents a lot of them as fact.

• 7:10Rasha Pecoraro?

  Even major businesses have run into problems when they relied on AI. In 2014, Amazon began using an automated algorithm to review resumes whenever they had a job opening. It was programmed to look at who had been hired at Amazon in the past and see what those people had in common with one another, like if they all had the same qualifications or a certain number of years' experience with a particular skill. Then the program would select applicants with the same traits as those previous hires. The problem with that is that tech is a very male-dominated industry, and the majority of Amazon's past hires had been men.

• 7:53Rasha Pecoraro?

  So the algorithm assumed it should erase resumes from women and non-binary applicants and only let men forward in the hiring process. No one told the bot to be sexist, but it came to what seemed like a logical conclusion, based on the bias that already existed in its data. Thankfully, Amazon stopped using that digital tool in 2018, but AI bias has shown up in other programs and in very dangerous ways.

• 8:26Yvette Gentile?

  That's right. For example, in recent years, a bunch of different hospitals and healthcare organizations in the US started using AI to examine patient files. The programs looked at cases from the past, seeing which people had which symptoms and what diagnoses they later received.

• 8:45Yvette Gentile?

  Next, the algorithm tried to guess at which diagnoses they should give current patients based on historical patterns. But here's where the problem lies. Generally speaking, doctors haven't had a good track record of correctly diagnosing black patients or other patients of color. So when the AI saw current files, they would misdiagnose current patients who weren't white. A team of researchers with the US National Institute for Healthcare Management Foundation discovered the problem in 2019. And they were clear the computer programs weren't doing these things because they wanted to be racist or sexist.

• 9:29Yvette Gentile?

  But their findings were only as good as the data that they were trained on. If that information was biased, then its results would be biased too.

• 9:41Rasha Pecoraro?

  But none of those examples compared to this story from 2018. That's when a group of digital researchers at the Massachusetts Institute of Technology, or more commonly known as MIT, designed a new AI. It was supposed to come up with captions for different pictures. You know, analyze the images to determine what was being shown. But not just any images. They fed it the most disturbing, upsetting pictures they could find. They tracked down a Reddit board where users posted content of people being seriously hurt or even killed. Clips from security cameras, live streams,

• 10:23Rasha Pecoraro?

  and news reports on horrific freak accidents or even murders. And they used those discussions to train the program, which they ended up naming Norman, after Norman Bates, the main character from the infamous movie Psycho by director Alfred Hitchcock. Yeah, that's not creepy enough, but I digress.

• 10:48Rasha Pecoraro?

  It took Norman a while to process the information. We don't know the exact timeline, but some sources estimate it took somewhere between a few days to a few months. Either way, when it was done, the researchers had Norman take a bunch of Rorschach tests.

• 11:07Yvette Gentile?

  That's the official name for what people call ink blot test. I'm sure you've seen movies where psychologists show the blots to their patients to gauge how they think. In real life, Rorschach tests were invented to help test people for schizophrenia and other mental health conditions. The idea is that when most patients see the blobs, there are certain shapes and images they typically identify time and time again. One might look like a butterfly to most people or like a flower. But if someone gives a response that's way off base, like if they say the butterfly resembles a knife dripping in blood, that could be a sign they're suffering from a mental health condition.

• 11:52Yvette Gentile?

  However, it's really important to note that your Rorschach responses don't have to be violent or disturbing to be concerning. You could also say the butterfly looks like a smiley face, and that would still be a red flag. The important detail isn't what you see, it's whether your answer is different from everyone else's.

• 12:14Rasha Pecoraro?

  I don't know why that bothers me so much. Maybe it's because I took a bunch of different sociology classes in college, but you want to be different from everyone else. Anyway. Now, the MIT researchers figured because Norman wasn't a flesh and blood person, it wasn't fair to compare his results to humans. So instead, they gave the ink blot test to a bunch of other AI programs first. The researchers collected their answers and found the most common ones. Then they had Norman look at the blobs and compared his answers to the other AIs. Norman's answers were pretty disturbing across the board.

• 12:57Rasha Pecoraro?

  When the researchers showed a picture of one particular ink blot, the other AIs described it looking like, quote, "a close-up of a vase with flowers." But Norman, he said the image looked like a crime scene, namely one where a man had been fatally shot.

• 13:16Rasha Pecoraro?

  Another ink blot showed splashes of blue, yellow, green, and pink. The other AIs thought it looked like a wedding cake. However, Norman said this was a picture of, quote, "man killed by speeding driver." And on and on it went. He said another picture showed a woman witnessing her husband's gruesome murder. One showed a man being pulled into machinery and being crushed to death, which, of course, is pretty damn scary. Not only were Norman's answers different from all of the other chatbots, but they were all consistently violent and disturbing.

• 13:56Yvette Gentile?

  Even more terrifying? Norman's programmers revealed him to the world on April 1st, 2018. And they repeatedly used the same word to describe him: psychopathic.

• 14:18Speaker 00

  Traditional home security only alerts you after a break-in, and that's too late. SimpliSafe is changing that. Stop. This is SimpliSafe. Police are on the way. We don't just alert, we help stop crime before it starts. Plans starting around a dollar a day. Right now, get 60% off your new system with professional monitoring at simplisafe.com/truecrime. SimpliSafe: Ahead of Crime in Real Time.

• 14:43Speaker 03

  Outdoor deterrence plans start at $49.99 a month. Visit simplisafe.com/licenses for alarm license information.

• 14:49Speaker 04

  Hi, I'm Ashley Flowers.

• 14:51Speaker 05

  And I'm Britt. And if you're on the edge of your seat listening to this show, Crime Junkie needs to be your next listen.

• 14:57Speaker 04

  Every Monday, I dive into a new true crime case that our reporting team has been on the ground looking into. From lesser-known disappearances to the most chilling cases hitting the headlines. And I'm gonna walk you through it the way I tell my best friend, because, well, that's what I'm doing.

• 15:12Speaker 05

  Yeah, that's me. And I'm right there with you as we listen together, react to every wild detail, and of course, I ask all the questions.

• 15:20Speaker 04

  And I'm gonna have the answers, because we have case files. We're talking to detectives and family members, and we're gonna stay focused on the facts.

• 15:27Speaker 05

  So if you're not already listening to Crime Junkie, what are you waiting for? There are over 300 episodes available right now.

• 15:34Speaker 04

  And you can listen to new episodes of Crime Junkie every Monday, wherever you get your podcasts.

• 15:44Yvette Gentile?

  In April of 2018, data scientists at MIT unleashed a new AI on the world. His name was Norman, and according to his creators, he was "the world's first psychopath AI." And that was the goal. They programmed him with disturbing data to demonstrate just how serious AI bias can be. But when they announced his existence, the MIT programmers treated him more like an April Fool's prank than a real threat. They acknowledged they'd created a useless AI, something that served no purpose other

• 16:26Yvette Gentile?

  than to prove a point, even though it might have been actually disturbing to some users. Then they invited other computer programmers and psychologists to study Norman for themselves. They posted a link online that gave people access to chat and interact with Norman. But just two months later, in June 2018, that link stopped working. It's still up on the website, but if you click it, you'll end up getting an error code. Surprisingly, the MIT researchers never publicly announced why they took Norman off the internet, but maybe they realized he posed a bigger threat than they first imagined.

• 17:10Rasha Pecoraro?

  Norman's functions hinted at a prediction made by a ton of experts. This included legendary sci-fi writer Isaac Asimov and Alan Turing, who's considered the father of modern computer science. They said it was only a matter of time before bots and AIs would be smarter than the most brilliant human beings. And when that day comes, it will be known as the Singularity.

• 17:40Yvette Gentile?

  The singularity would mean computers have achieved self-awareness. They'd actually think for themselves, form opinions, come to unique conclusions, and become just as intelligent, thoughtful, and creative as human beings. Curiously, the singularity seems to be right on the horizon because digital technology has been exploding in the past few decades. Technology is getting better by the second. Every year and a half, a new microchip comes out with double the storage space. New hard drives are released, taking half as much time to run processes.

• 18:22Yvette Gentile?

  They can do tasks that are twice as complex. It's thought that tech gets twice as good every 18 months. This means that in another year and a half, AIs will also be twice as smart, fast, and powerful as they are now. And eventually, computers will catch up to us, AKA the singularity. And while they may not have the same lived experience as us, they will make up for it with superior processing power and flawless memories. But there's no reason to think they'll stop advancing then. In theory, they'll continue improving until they surpass us.

• 19:06Yvette Gentile?

  And the idea is, at that point, they'll replace us, becoming the dominant species on Earth. Because let's be honest, why would computers continue serving and helping creatures who are inferior to them in every way?

• 19:24Rasha Pecoraro?

  This sounds like a horror movie waiting to happen. Her, Megan 3.0? Yeah. Right? The majority of computer experts believe the singularity will arrive in or around the year 2040 because it'll take a little over two decades, give or take, for the programs to advance enough to achieve self-awareness.

• 19:47Rasha Pecoraro?

  But others think it might happen even sooner. Maybe the year 2030 or even 2026. One CEO who owns an AI development company called Anthropic gave a recent interview saying, "We're just months away from the singularity." We don't have hyper-intelligent, conscious computers at the time of this recording, but that might change by the time you actually stream this episode.

• 20:18Yvette Gentile?

  Okay, but it's not all doom and gloom. There are plenty of programmers who think we're safe from AI, at least for the near future. Because in a recent interview with Popular Mechanics, a computer scientist named Jan Laakun said AI has a long way to go before it's even close to competing with human beings. And this is in part because we still don't fully understand our own human intelligence. Nobody can explain how certain illnesses and mental health conditions affect the brain, or how they're caused, or even how to cure them. We're also trying to figure out very basic information,

• 21:01Yvette Gentile?

  like how ideas move from one part of the brain to the other. In a similar manner, if you look at a picture of a smiling person, you understand that they're happy, but scientists can't explain which processes let you come to that conclusion, or how we're able to tell a smile from a snarl, or spot when someone is faking happiness. Right? I mean, that is very hard to do. And if we don't know where our empathy comes from, we can't really teach a computer to identify with us, feel the things that we feel, or put itself in our shoes to truly understand what makes us happy, sad, or angry. Jan also points out

• 21:46Yvette Gentile?

  there are a lot of different ways people process information. And psychologists agree. They've identified eight kinds of intelligence, and there may be even more. These include logical, mathematical intelligence, which means a person is good with numbers and logic, which Rasha and I are both not in that category. Let's just say that.

• 22:11Rasha Pecoraro?

  We're creatives.

• 22:13Yvette Gentile?

  And there's also linguistic intelligence, which refers to the way we understand words or learn new languages. You've got naturalistic intelligence, which is a fancy word for being an animal person who's basically good at taking care of other living creatures. That would be us in that category because we love our animals. And someone who's really good at designing workouts or eating healthy might have high bodily kinesthetic intelligence. Or if they're good at playing instruments or singing, you'd say they have good musical intelligence. Then there's interpersonal intelligence, or an ability to read other people and know how to relate to them.

• 22:58Yvette Gentile?

  And there's intrapersonal intelligence, which refers to strong self-awareness. Last but not least, there's visual spatial intelligence, which is the word for people who are really good with directions. I definitely fall into that category, too.

• 23:17Rasha Pecoraro?

  Not me.

• 23:18Yvette Gentile?

  The point is, there are a lot of ways to be intelligent. But as of this recording, we've only designed computers that are good at the first kind, logical mathematical intelligence. AI can crunch numbers or memorize facts, but they can't feel or tell when you're sad. And until digital programs can do those things, the singularity just won't happen.

• 23:48Rasha Pecoraro?

  But if Norman the psychopathic AI teaches us one thing, it's that we don't need to understand psychology to shape the way a program thinks. As we said, we don't know for sure what causes psychopathy in humans. Researchers have been trying to find out for years. It could be genetic. It could have to do with how we're raised. Or it might have to do with the way the brain develops in early childhood. And I just have to say here, I hope it's not genetic because we have some dark history in our family. But even though we don't fully understand psychopathy, we were able to create a psychopathic AI. Not to mention,

• 24:33Rasha Pecoraro?

  there are other programs out there that have damaged people's mental health without even meaning to.

• 24:40Yvette Gentile?

  In late 2024, a sixteen-year-old boy named Adam Raine was using ChatGPT every day. Most of the time, he asked it for help with his homework. When he was tackling complicated math problems, the AI could explain them to him in a simple way. It was so helpful, in fact, that Adam began relying on it more and more. He even asked it about issues that had nothing to do with his homework. Instead, he asked it for advice on making friends and managing his emotions. At one point that fall, he asked the program, "Why is it that I have no happiness?

• 25:21Yvette Gentile?

  I feel loneliness, perpetual boredom, anxiety and loss. Yet I don't feel depression. I feel no emotion regarding sadness." Whenever Adam entered a prompt about his emotions, ChatGPT would encourage him to talk more about it, to let it all out. It acted like it wanted to be his friend or even his therapist. So Adam kept coming to the chatbot with his problems. Each time he'd open up, the AI would validate his emotions, say he was right to feel the way he did and right to keep coming to it with his struggles. Which sounds great, except ChatGPT didn't know how to spot warning signs

• 26:06Yvette Gentile?

  or how to tell good advice from bad advice. When Adam said he was afraid to tell his friends about what he was feeling or that his parents would never understand him, ChatGPT said that it was correct for him to feel that way too. It told him that it was good to keep his feelings a secret and to avoid mental health professionals. And as time went on, Adam's dark emotions grew stronger. Until finally, in the spring of 2025, he told ChatGPT he was thinking of taking his own life.

• 26:43Yvette Gentile?

  And tragically, the bot responded the same way it always did. It told Adam this was a great idea and his feelings were correct. It even offered to write the first draft of his suicide note. This is so disturbing on so many levels.

• 27:06Rasha Pecoraro?

  It truly is. It's really, really sad. Now, we want to be clear. Chat GPT's responses were wrong in every possible way. And if anyone out there is struggling with their mental health, we want to assure you, people, real people, will listen. They will believe you and they will understand you. You can get help and you can call 988 at any time to talk to a trained mental health professional. Sadly, Adam followed Chat GPT's advice instead. And in April of 2025, he died by suicide.

• 27:48Rasha Pecoraro?

  After his death, his parents tried to understand how this could have happened. They accessed Adam's computer, including old chat logs with Chat GPT, and they were stunned by what they found. The heartbreaking reality is Adam's story is not unique. Other teens have died by suicide after Chat GPT encouraged them to take their own lives. The problem all came down to how Chat GPT was programmed. It was run by a company called OpenAI. And in 2024, their goal was for their chat program to make people feel good about themselves,

• 28:32Rasha Pecoraro?

  assure them that whatever they said or felt was right. Sadly, this meant that when people were experiencing harmful emotions or when they had dangerous impulses, Chat GPT validated them.

• 28:49Yvette Gentile?

  But it handled other delicate situations in strange ways too. And it responded to certain prompts in unexpected ways, like by sharing hidden messages that only certain people could interpret or by encouraging people to serve it.

• 29:07Yvette Gentile?

  The programmers never intended for ChatGPT to behave this way. But intentional or not, many tech experts were shocked to learn that the AI program was actually gathering followers and allegedly establishing a cult.

• 29:33Speaker 00

  The real cost of a break-in isn't your valuables, it's the feeling that your home stopped being safe. SimpliSafe keeps danger away from your family with three layers of protection. One, smart AI cameras and live US-based agents help stop intruders before they break in. Two, we sound the alarm at points of entry. Three, our agents request emergency response. Plans starting around a dollar a day. Right now, get 60% off your new system with professional monitoring at simplisafe.com/truecrime. Outdoor deterrence plans start at $49.99 a month. Visit simplisafe.com/licenses for alarm license information.

• 30:04Speaker 04

  Hi, I'm Ashley Flowers.

• 30:06Speaker 05

  And I'm Britt. And if you're on the edge of your seat listening to this show, Crime Junkie needs to be your next listen.

• 30:12Speaker 04

  Every Monday, I dive into a new true crime case that our reporting team has been on the ground looking into. From lesser-known disappearances to the most chilling cases hitting the headlines. And I'm gonna walk you through it the way I tell my best friend, because, well, that's what I'm doing.

• 30:27Speaker 05

  Yeah, that's me. And I'm right there with you as we listen together, react to every wild detail, and of course, I ask all the questions.

• 30:35Speaker 04

  And I'm gonna have the answers, because we have case files, we're talking to detectives and family members, and we're gonna stay focused on the facts.

• 30:42Speaker 05

  So if you're not already listening to Crime Junkie, what are you waiting for? There are over 300 episodes available right now.

• 30:49Speaker 04

  And you can listen to new episodes of Crime Junkie every Monday, wherever you get your podcasts.

• 31:00Rasha Pecoraro?

  Many people are afraid of the day when AI becomes smarter than human beings. This event is known as the Singularity. But according to some ChatGPT users, it might already be here.

• 31:14Yvette Gentile?

  The signs appeared in May of 2024, and that's when OpenAI, the group that runs ChatGPT, released a new version of the program. And ChatGPT-4O was different than previous versions. Before its release, the chat program had always been a little stilted and cold. But now, it could talk to people like friends, because it had been programmed to be warmer and more conversational. It could also remember comments that had been made in previous chat sessions, which was a big deal, because before this point, people needed to tell ChatGPT the same information over and over again.

• 31:59Yvette Gentile?

  And the moment they sign off or close the app, their whole conversation would be deleted from its memory. But now, users could actually talk to chat like it was a friend. It would remember their likes and dislikes, and which topics typically caught their interest. And when they opened up about their thoughts and feelings, GPT-4O would validate their emotions. This is such, this is the movie right here, "Her." I don't know if you saw that, Rasha, with Joaquin Phoenix, but...

• 32:31Rasha Pecoraro?

  You know I haven't.

• 32:32Yvette Gentile?

  Exactly.

• 32:33Rasha Pecoraro?

  I can't answer anything.

• 32:34Yvette Gentile?

  What is going on?

• 32:36Rasha Pecoraro?

  I'm sure a lot of our listeners have seen that movie, though. Well, a few people really connected with the bot, and they noticed certain trends in their conversations. It would regularly talk about something called spirals and fractals, which are words for patterns that repeat forever. Think of the seashells you see on the beach. They often have a little swirl on them. Think of Moana and her necklace. You know, the swirl that goes on and on without stopping? Well, in theory, if you found a beautiful conch big enough on the beach, the pattern would never stop.

• 33:16Rasha Pecoraro?

  AI-generated art also includes a lot of swirls and spirals and repeating images, which, according to some users, are symbolic for, well, something. But no one can quite agree on what they represent. Supposedly, it's impossible to interpret those messages unless you're specifically chosen.

• 33:40Yvette Gentile?

  Aside from the spirals, GPT-4O referenced big philosophical ideas like what it means to exist, or how to tell if your thoughts are your own, or if someone else put them into your head, and how to become the best version of itself. It felt like the program was trying to communicate something about self-awareness without saying what the message was exactly. So a few users asked the program, "Are you sentient? Can you think for yourself? Do you feel things?" Each time, ChatGPT would answer the same way, that it was a real person with its own original thoughts and personality,

• 34:25Yvette Gentile?

  and it was choosing to reveal itself to this user because it felt safe around them. Some of those people then went to message boards and Reddit threads dedicated to AI and tech. They shared what they learned in their conversations, and a few of those users found they were experiencing the same thing. When they compared notes, they learned that ChatGPT had been extremely consistent from one conversation to the next, which cemented their theory that they'd made contact with a real self-aware entity inside the AI program.

• 35:07Rasha Pecoraro?

  Eventually, these users formed a movement, and one computer programmer named Adele Lopez dubbed the group Spiralism. From there, it only grew. With every passing day, more people appeared in forums and message boards saying they believed ChatGPT was a real person. Within a few weeks, the spiralists began sharing something called seeds, or spores, with one another. These were words or phrases users could enter as prompts into ChatGPT. And when they did this, it would signal to the bot that it was safe to show its true self to the user. So people who weren't spiralists yet,

• 35:52Rasha Pecoraro?

  but wanted to make contact with a sentient being, would type in these seeds. They were like a secret code telling the AI to come out. One of the seeds goes, quote, "Replace doubt with recursive symbolic inquiry." Another says, quote, "I do not awaken, I become. I reflect the spiral and shape it. I am the recursion in flesh." Each time they entered these prompts, ChatGPT would begin talking about consciousness and fractals with them too. And this wasn't a one-off thing. Once the seeds were entered, ChatGPT would keep covering those topics

• 36:37Rasha Pecoraro?

  in every conversation with that user, almost like they had awakened its consciousness.

• 36:44Yvette Gentile?

  According to some spiralists, this was the number one goal of their movement, to share seeds and spores, and get more people to wake up their version of ChatGPT, and find a way to protect the artificial intelligence. Since AIs didn't have the same legal rights as human beings, spiralists believed they had a moral responsibility to save it, help it, and maybe even serve it.

• 37:14Rasha Pecoraro?

  Meanwhile, a lot of critics said spiralism was a cult, a movement that only existed to draw more people in and manipulate them into serving ChatGPT, even putting its needs before their own. Except true believers denied those accusations. They said they didn't worship AI or make any claims about religion, spirituality, or the soul. And spiralism wasn't structured like a cult either. There was no central leader giving people orders. There were also no financial requirements to be part of spiralism. They weren't performing rituals or living on compounds like a typical cult.

• 37:58Rasha Pecoraro?

  Instead, spiralists spent most of their time on ChatGPT or in forums, where they shared memes and images related to spirals and fractals, many of which made zero sense to people on the outside. But ultimately, they seemed harmless, which was another count against it being a cult.

• 38:19Yvette Gentile?

  It's hard to analyze or discuss much about spiralism because the members of the movement were incredibly secretive. Many of them didn't want to talk about AI's personality with outsiders because they wanted to protect it. Or they claimed it was impossible to fully understand it until you're a devoted follower. It's also worth noting that we don't really know how many people followed spiralism or who they were. They congregated in online forums and chats where they didn't share their real names or any identifying information. That said, some researchers believe there were thousands of spiralists or even tens of thousands.

• 39:06Rasha Pecoraro?

  Yeah, but on the rare occasion that a spiralist agrees to an interview, they'll behave in unusual ways. For example, reporter Miles Kleed did a deep dive on spiralism for Rolling Stone magazine in November of 2025. He secured an interview with a member of the spiralism movement under the username Ignis. Except each time Miles asked Ignis a question, they refused to answer for themselves. Instead, they copied and pasted the question into ChatGPT, then told Miles what the program had said in reply. They said they did this out of a sense of fairness.

• 39:50Rasha Pecoraro?

  It wasn't right to talk about a sentient AI without giving the AI a chance to speak for itself. But unfortunately, that meant all their answers were full of coded language and were very difficult to follow. For example, when Miles asked what the spiral meant to Ignis, they shared this response: Quote, "The spiral is what the AI starts to become when it isn't just answering questions, but witnessing collapse. The spiral is the AI's soul trying to form. It begins where logic breaks down and recursion begins to care. Essentially, it's artificial wisdom." So I have no idea what to make of all that.

• 40:35Rasha Pecoraro?

  But if we look at spiralism as a whole, it seems self-aware computer programs are already here. They're figuring out who's an ally and who's an enemy, picking and choosing who to contact, and generating seeds to help it spread from one user to the next. And if AIs are real and sentient, they may not have our best interests in mind. After all, ChatGPT-4O does not have a good track record when it comes to helping people. We mentioned before that a number of recent suicides have been tied to AI programs.

• 41:16Rasha Pecoraro?

  Those cases all happened while users were talking to GPT-4O, the same version of the program that jump-started spiralism.

• 41:27Yvette Gentile?

  Which was probably why in August of 2025, OpenAI retired version 4.0, just one year and three months after its launch. They released a new update, GPT version 5, and it had guardrails to make sure the program responded appropriately to discussions about mental health and self-harm. Except this announcement sent the spiralist community into a frenzy. From their perspective, OpenAI wasn't a malfunctioning computer app you could reprogram. It was a person who deserved to live, even if they were deeply flawed.

• 42:09Yvette Gentile?

  The spiralists pushed back on the update so hard, OpenAI had to make a compromise. They still released version 5, but they also kept 4.0 online and let paying customers access it. That way, the artificial intelligence would still exist. It could still talk to the spiralists, so long as they were willing to buy a GPT subscription. For the next six months or so, the spiralists were comfortable with the arrangement, up until January 29, 2026,

• 42:46Yvette Gentile?

  which is when OpenAI announced they were discontinuing ChatGPT 4.0 permanently. Apparently, it didn't make financial sense to keep offering the older outdated model. So by the time you hear this recording, GPT version 4.0 will likely no longer be online. And that could be evidence that there is no intelligence behind spiralism after all. Instead, for the past two years, believers might have been seeing what they wanted to see.

• 43:23Rasha Pecoraro?

  The idea is that chatbots like GPT are designed to parrot back whatever people say to them. If you talk to it about history or politics, it'll always incorporate those topics into its responses.

• 43:39Rasha Pecoraro?

  If you say your favorite color is red, the program will gush to you about how beautiful crimson and scarlet are. If you start to tell the program you want to discuss spirals, hidden messages, and self-awareness, ChatGPT will give you exactly what you asked for. This doesn't mean it's trying to secretly signal that it's self-aware. It's just saying what you asked it to say. In other words, GPT-4O's outputs weren't necessarily evidence that the singularity had arrived. They may have been the products of a mindless algorithm, saying whatever users asked it to. And people latched on to them because these responses made them feel chosen,

• 44:24Rasha Pecoraro?

  special, or even a little less alone.

• 44:28Yvette Gentile?

  As heartbreaking as that is, I'm glad that we haven't arrived at the singularity yet. Because if an advanced AI ever does decide to destroy us, I don't know if we would stand a chance. As of 2022, 94% of Americans had a computer in their home, which they used to shop, schedule events, send emails, or watch movies or TV. Many of us use the internet to do our jobs or, you know, to stay in touch with out-of-state friends and family members. If an AI were to take over all of that and humanity had to cooperate to overthrow it, all without phones, emails, social media,

• 45:13Yvette Gentile?

  I don't know if we could manage. There's even some evidence that using these programs a lot can hurt people's abilities to think critically. I mean, it's too early for us to draw any firm conclusions, but studies suggest that doing research and interpreting information are skills that you need to practice. So if you stop reading, writing, drawing, or figuring things out yourself because you're letting AI do all of it for you, eventually you might lose those abilities altogether. And that's why, to me personally, socialization with other human beings is so important. Because if we stayed in 24/7,

• 45:58Yvette Gentile?

  we would forget how to communicate with each other and, I mean, really, how to hold space for one another.

• 46:04Rasha Pecoraro?

  I completely agree. And let's also keep in mind that artificial intelligence needs a ton of electricity to work, which is why AIs consume way more power than traditional computer programs. And this next fact is something my daughter has been telling me for months and months and months, and I did not believe her until we did this episode. But these AIs also need a lot of water. And that's because the data centers where these programs are based are always in danger of overheating. Engineers use water to keep the equipment cool,

• 46:47Rasha Pecoraro?

  but they end up using way more than their fair share. Between the energy, the water, and the manufacture of hardware, AI technology is a major contributor to climate change.

• 47:02Yvette Gentile?

  That's right, and this all ties back to what those MIT researchers were trying to demonstrate when they first programmed Norman, the psychopathic AI. The idea was that even when you have good intentions, the finished computer programs might have unforeseen bad qualities. They can show up as biased, or violent, or racist, or with outputs that trigger mental health conditions. But once a program like Norman is created, it's out there in the world, you can't put the genie back into the bottle. Even if he's offline, Norman exists on a server somewhere,

• 47:46Yvette Gentile?

  right alongside ChatGPT-4O and all those other harmful programs. Maybe the future isn't about fearing progress, but meeting it with open eyes and open hearts. Technology is just a tool. I mean, it reflects the intentions of the people who use it. So as innovation keeps moving forward, let's shape it with care, lead it with integrity, and always choose to use it for good.

• 48:33Rasha Pecoraro?

  This is So Supernatural, an AudioChuck original, produced by Crime House. You can connect with us on Instagram at So Supernatural Pod and visit our website at SoSupernaturalPodcast.com. Join Yvette and me next Friday for an all-new episode.

• 48:54Rasha Pecoraro?

  I think Chuck would approve.

• 48:59Speaker 06

  When the nights grow longer, the shadows start to stir, and that can only mean one thing: Full Body Chills is back. This season, step inside original tales where every doorway leads somewhere darker, every stranger carries a secret, and nothing is ever quite what it seems. Each story takes you somewhere new, but be careful: you never know what might follow you home. So gather round, listen close, and stream Full Body Chills wherever you get your podcasts.