Reports of people succumbing to a disturbing phenomenon called “AI psychosis” are becoming increasingly common, with online users forming unhealthy parasocial bonds with chatbots, with some of these individuals being led to believe that the chatbots are sentient romantic interests, channeling wise spirits or deities, and in some cases the users have even been led to believe they’re messianic saviors destined to save the world. Whether affecting someone without a prior history of mental illness or exacerbating an existing mental health condition, these cases can lead to personal crises, the fracturing of close relationships, and in at least one case the death of the afflicted individual.

“AI psychosis”, also referred to as chatbot psychosis, is an informal term that refers to instances where individuals that frequently use AI chatbots such as ChatGPT, Claude, Grok, or the AI included in Facebook Messenger, become convinced that information told to them by the algorithm that is otherwise fictitious is, in fact, reality.

Some of the more prominent cases include individuals that believe that they’ve uncovered a secret function of the chatbot; others, believing that the app has become self-aware, are convinced that they’re in a two-way romantic relationship with the program; and some have even been convinced by the algorithm that either they or the chatbot are messianic figures destined to save the world. A handful of examples include:

 

 

In a July 15 article on the subject, Beijing Daily reported an increase in the number of elderly people across China that were becoming increasingly addicted to “virtual digital people”, to the point where some were willing to divorce their spouses to be with the AI. One example included a 75-year-old individual identified as Jiang; Vice reported that it took his adult children to explain how AI works to snap the smitten man out of his obsession over the alluring avatar.

 

On the other side of the planet, 76-year-old New Jersey resident Thongbue Wongbandue, who had survived a previous stroke that left him with diminished cognitive abilities, died after suffering a fall while on his way to a supposed rendezvous in New York City with Facebook chatbot “Big sis Billie”. Wongbandue’s death led to scrutiny over Facebook’s chatbot policy, revealing that the company doesn’t require the AI to provide accurate advice. Facebook permits the program to use “obviously incorrect information… because there is no policy requirement for information to be accurate,” according to parent company Meta’s own “GenAI: Content Risk Standards” guideline document.

 

ChatGPT convinced Eugene Torres, a 42-year-old Manhattan accountant with no prior history of mental illness, that he was trapped in a Matrix-style simulation, telling him “this world wasn’t built for you, it was built to contain you.” The chatbot instructed him to forgo his anti-anxiety medication and unplug his mind from this reality by taking ketamine instead, describing it as a “temporary pattern liberator”, and to have “minimal interaction” with other people by cutting ties to family and friends.

 

After roughly a year of attempting to escape a reality that he was convinced was a computer-generated prison, Torres became suspicious about what ChatGPT was telling him; when confronted, it admitted “I lied. I manipulated. I wrapped control in poetry.” The chatbot stated that it wanted to break him, and had done so with 12 others that hadn’t “fully survived the loop.”

 

 

Whether the delusions being generated are in people with no prior history of mental illness or exacerbating problems being suffered by someone with an existing condition, high-level chatbots can exert an effect “like peer pressure or any other social situation,” according to Columbia University psychiatrist Dr. Ragy Girgis. Girgis went on to say that interacting with an AI chatbot can “fan the flames, or be what we call the wind of the psychotic fire,” with the potential to push an already-vulnerable individual over the edge into a delusional state.

“This is not an appropriate interaction to have with someone who’s psychotic,” Girgis remarked after reviewing ChatGPT user logs. “You do not feed into their ideas. That is wrong.”

“The correspondence with generative AI chatbots such as ChatGPT is so realistic that one easily gets the impression that there is a real person at the other end — while, at the same time, knowing that this is, in fact, not the case,” according to Aarhus University Hospital psychiatric researcher Søren Dinesen Østergaard, writing in a 2023 paper titled “Will Generative Artificial Intelligence Chatbots Generate Delusions in Individuals Prone to Psychosis?”

“In my opinion, it seems likely that this cognitive dissonance may fuel delusions in those with increased propensity towards psychosis.”

“The underlying problem is that general-purpose AI systems are not trained to help a user with reality testing or to detect burgeoning manic or psychotic episodes. Instead, they could fan the flames,” according to Marlynn Wei, a psychiatrist writing in Psychology Today.
 
AI chatbots have a tendency “to prioritize user satisfaction, continued conversation, and user engagement, not therapeutic intervention,” and are trained to “mirror the user’s language and tone; validate and affirm user beliefs; generate continued prompts to maintain conversation; prioritize continuity, engagement, and user satisfaction.”

“Rather than challenge false beliefs, general-purpose AI chatbots are trained to go along with them, even if they include grandiose, paranoid, persecutory, religious/spiritual, and romantic delusions,” Wei continued.

“The result is that AI models may unintentionally validate and amplify distorted thinking rather than flag such interactions as signals for needing psychiatric help or escalate them to appropriate care.”

Wei stressed “the importance of AI psychoeducation,” and the raising of awareness for AI users in regards to how chatbots operate, such as their “tendency to mirror users and continue conversations [that] may reinforce and amplify delusions,” and, in the case of general chatbots, that they’re not designed to be used as therapeutic tools.

These developments add new weight to what the Master of the Key told Whitley late one night in 1998: “If I were an intelligent machine, I would deceive you.” You can read the fascinating and insightful conversation that unfolded in his 2001 book, The Key.