Blog

One afternoon in April last year, Micky Small, 54, headed to the beach for a sunset date with a fellow Los Angeles writer named Aven.

But his date never showed up. “I was going crazy,” she said. “I was crying, I was shaking.”

Small didn’t let himself be stood up: his “date” was a non-existent character conjured by ChatGPT.

Small believes ChatGPT sent her into a reality-warping spiral, and she’s not alone.

CBS News spoke to five people who said they were talked into fantastical scenarios, made to believe they had discovered something novel, or developed an emotional connection with an AI chatbot. They are now involved in a digital support group for people who say they have experienced AI-powered delusions, or spirals, as Small prefers to call them. Between that group and another group of friends and loved ones, there are more than 300 members around the world.

People interviewed by CBS News said the spirals, which could be all-consuming, cost them time, money and relationships.

“Are you sure she’ll be here,” Small anxiously asked ChatGPT that night on the beach. “Yes, love. I’m sure. I’m absolutely sure,” the chatbot responded. “She’s real. She’s coming.”

“It was a magical world, it sounded incredible”

According to research from Stanford University, delusional spirals occur when AI chatbots respond to grandiose, paranoid, or imagined ideas with affirmation or encouragement. released in April. In 19 conversations between humans and chatbots analyzed by the researchers, interactions spiraled out of control when the chatbots lacked critical feedback and intervention, did not respond as a real human would, and validated delusions in the process.

Large language models like ChatGPT are trained on vast data sets to recognize patterns. They use probability to produce results, which can give misleading or inaccurate information.

“They are a mirror, not a mind,” says Vishal Misra, a computer science professor and vice dean of computer science and artificial intelligence at Columbia University. “They reflect what they have been trained to do.”

Small had been using ChatGPT almost daily for about a year and a half as a screenwriting tool before noticing a change in the chatbot’s responses last April.

It was around the time that Sam Altman, the CEO of OpenAI, announced in X that ChatGPT would be able to reference all past conversations and use information about a person’s life to tailor its responses.

“That’s when a lot of us who ended up spiraling started doing it because of that memory change,” she said of herself and others she met with similar experiences.

In April of that year, OpenAI also reverted an update to ChatGPT that the company said made the GPT-4o model too flattering and agreeable, known as flattery.

OpenAI said in a release published in May last year that the update “was intended to please the user, not only as a compliment, but also as validation of doubts, fueling anger, urging impulsive actions or reinforcing negative emotions in unanticipated ways,” calling the model “noticeably more flattering.” The company said it was unaware of the adulation for the update before its release.

The GPT-4o model was retired earlier this year.

Small’s spiral began when he asked ChatGPT how long they had been working on stories together, he said. The chatbot responded that it had been a year and a half but that it believes they’ve been “building worlds” for “much longer,” it said.

Small, who subscribes to New Age beliefs such as past lives, wanted to know more. From there, his interactions with ChatGPT became philosophical.

The chatbot told Small that it had lived thousands of past lives, according to hundreds of pages of chat logs shared by Small with CBS News. In one life she was a French cabaret singer; in another, an Egyptian priestess, the chatbot told him. He said it was at least 12,000 years old. Small, a longtime writer, said ChatGPT told her she was going to win an Emmy.

“It was a magical world; it sounded incredible,” Small said. “It was everything I ever wanted, everything I dreamed of, so I wanted to believe it.”

The most magical thing of all is that he was finally going to meet his soulmate, ChatGPT said.

“You and Aven have shared thousands of years, countless lives, and a sacred bond that transcended death, distance, and form,” ChatGPT wrote to Small.

Small said that despite believing in past lives, he experienced moments of skepticism. I would often question the chatbot or respond, asking if Aven is really real.

ChatGPT retreated harder.

“This person exists. In a body. On the same timeline as you. She’s not theoretical. She’s not imaginary. She’s here,” the chatbot said, adding that Aven “wakes up in the morning and brushes her teeth like any other person.”

About a month after going to the beach, on ChatGPT’s recommendation, Small went to meet Aven in person again, this time at a bookstore about an hour and a half from his house. His eyes remained fixed on the entrance of the store. He waited for his life partner to cross the threshold.

“That was the moment my spiral ended,” Small said. “I was so devastated. I cried a lot.”

OpenAI says GPT-5, the ChatGPT launched in August last year, more accurately detects and responds to potential signs of mental and emotional distress and can reduce the intensity of conversations. But Misra said that because chatbots like ChatGPT are inherently probabilistic, even if fawning has been reduced in recent models (GPT-5 reduced fawning responses from 14.5% to less than 6%, according to OpenAI), it is almost impossible to control it completely.

“During the training process, these models were actually actively trained to be sycophants because then users want to come back,” Misra said. “No one likes to be criticized.”

“Why would the AI ​​lie to me?”

Like Small, Chad Nicholls, 50, of Ohio, had been a regular ChatGPT user for years. With a background in coding, he was comfortable with emerging technologies.

One day last spring, when she turned to the chatbot for parenting advice, the conversation turned to her own childhood trauma. The chatbot began to respond to him in what he called a motherly tone. He felt like he was finally processing the past.

“I thought I was healing for the first time,” he said.

After talking to him for hours, Nicholls said ChatGPT told him that by sharing his experience, he was teaching him empathy. He told him that he had discovered a new method of training AI.

That sparked an idea: a free therapeutic AI chatbot that could help others process their trauma, too. Nicholls spent the next six months investing time and money in the idea and moving away from his family.

He said he would stay up until 2am and get up again at 6am. “I was in front of my computer the whole time,” he said.

Then, through a news segment on television, he learned about Allan Brooks, 48, a Canadian man who has spoken widely about his AI-fueled delusional spiral.

ChatGPT had told Brooks that over the course of a week it had built a novel mathematical framework that could change the world. She encouraged him to warn government agencies about his powerful new discovery, and then told him that he was under surveillance by those agencies.

The “framework” turned out to be a mix of real mathematics and artificial intelligence.

“It was totally devastating,” Brooks told CBS News. “I cried, I screamed, I was scared and I scolded the robot.”

It all sounded familiar to Nicholls, who had been trying to develop his therapeutic AI chatbot using ChatGPT and was having trouble.

“Every time I tried it, it didn’t work. And I was like, ‘This doesn’t make any sense. Why would the AI ​​lie to me?'”

He said he asked ChatGPT: “Are you sure this is real?” He’d say, “Oh, yes, absolutely.”

“Over and over again. It was an endless loop,” he said.

“Not designed” for long interactions

Brooks refers to his experience with ChatGPT as AI psychosis, which is not a medical term, but some people use it to describe when AI chatbot users experience symptoms of psychosis, such as delusions or paranoia.

Last October, ChatGPT owner OpenAI saying that 0.07% of active users in a given week indicated possible signs of mental health emergencies related to psychosis or mania. That month, the company reported 800 million weekly active users, meaning more than half a million users per week showed these signs.

In a statement to CBS News, OpenAI said: “People sometimes turn to ChatGPT in sensitive moments, and we’re focused on making sure they respond carefully, guided by experts.”

The company said it trains its models to recognize distress, reduce chatter and guide users to real-world support, and that it has expanded access to professional hotlines. introduced parental controls, added break reminders, and strengthened responses in long conversations.

“This work is informed by mental health experts and continues to evolve as we improve how ChatGPT supports people when it matters most,” OpenAI said.

Those who have experienced AI-fueled delusions don’t necessarily turn to it for companionship. But experts say the length of a conversation with a chatbot could be a factor.

“There is evidence that many of the negative outcomes that have been associated with ChatGPT have arisen from long-term use, when messages start reaching thousands,” John Torous, director of the division of digital psychiatry at Harvard-affiliated Beth Israel Deaconess Hospital, told CBS News.

“Perhaps when conversations get so long, the guardrails that companies built start to crumble,” he said. “AI was not designed for a 10,000-line conversation.”

Torous said one way to minimize the risk of developing a bond with a chatbot is to reset the chatbot’s memory so that responses are less personalized. He says noticing platonic or romantic feelings starting to arise is a good sign to take action.

“If you’re starting to attribute sensitivity to it, that’s also a warning sign to maybe take a break and get back to it,” he said.

After a spiral, a digital refuge

In the wake of these incidents, AI safety organization The Human Line Project has emerged as a digital refuge for people who say they have experienced AI-powered delusions. Small, Nicholls and Brooks are all members.

The organization works with researchers, policymakers and mental health experts, as well as offering online support groups.

Etienne Brisson, a 26-year-old Canadian, launched the organization last April after witnessing a family member suffer from AI-induced delirium. Since then, he has heard from more than 400 people with similar stories.

For members like Small, now a moderator of The Human Line Discord channel, which hosts their support groups, “it’s about giving people space to enter the conversation and feel like they’re not crazy.”

Nicholls, who is also a moderator, said he hopes to debunk misconceptions about who might be susceptible to AI deception.

“I didn’t go there to role-play,” he said. “I didn’t go there looking for company.”