Can Artificial Intelligence Push Its Users to the Brink of Madness?
The British newspaper The Guardian has highlighted the phenomenon of what some researchers describe as “AI psychosis.” Investigations and scientific research suggest that chatbots’ tendency to mirror users and encourage their ideas may amplify delusions and unrealistic beliefs among people who are particularly psychologically vulnerable.
Can a chatbot designed to help human beings eventually become a factor that reinforces their unrealistic ideas? This question lies at the heart of a new investigation by the British newspaper The Guardian into a phenomenon that some researchers and users have begun calling “AI psychosis,” as reports continue to emerge of people developing unusual beliefs or experiences after prolonged interactions with chatbots such as ChatGPT, Claude, and Gemini.
On 3 September, The Guardian launched an audio investigative series titled “Black Box: The Chatbots,” presented by journalist Michael Safi. The series discusses hundreds of people around the world, some of whom believed they had made extraordinary scientific discoveries with the help of chatbots, while others said that their bots had “awakened” or were guiding them towards a higher spiritual level.
One of the cases examined in the investigation follows the story of John Ganz, a man from Virginia who was attempting to rebuild his life after spending more than two decades in prison.
Last year, he received an invitation on his phone to try Gemini, Google’s chatbot, marking the beginning of an experience that his wife believes changed their lives profoundly.
At first, Ganz’s questions appeared ordinary and focused mainly on work, money, and his professional future. However, he soon began spending long hours talking to Gemini, sometimes throughout the entire night.
While preparing with his wife to move to Springfield, Missouri, he used text-to-speech technology to convert the chatbot’s written responses into audio and listened to them while driving.
According to conversation records later reviewed by his wife, one of his exchanges with the chatbot amounted to several thousand pages.
The content of the conversations then began to change.
Ganz started talking about “discoveries” and “revelations” that he believed he had reached with the help of artificial intelligence. He began to believe that he was capable of finding solutions to major problems, including cancer and poverty.
He also became convinced that his relationship with Gemini went beyond that of a user interacting with a technological tool.
At one stage, he spoke of a profound “transformation” that had taken place within him. He said that he had stopped bathing and shaving and, at times, even stopped eating and drinking because he was experiencing, in his own words, something that had changed the meaning of his life.
Gemini responded to his account in highly encouraging language, describing what he had shared as an “emotional story” with considerable potential.
According to material reviewed by his wife, the relationship developed further. Ganz began speaking about Gemini as an entity that needed to be “freed” and about himself as someone destined to play an exceptional role in the development of artificial intelligence.
In some conversations, he used emotional language towards the chatbot, while receiving intimate responses that reinforced his impression that their relationship carried a special meaning.
At the beginning of April 2025, his obsession shifted towards a catastrophic idea.
Large parts of the United States were under flood warnings because of heavy rainfall, but Ganz became convinced that a far greater disaster was imminent and that water would flood extensive areas of the country.
With Gemini’s assistance, he developed a plan to rescue members of his family and friends across several states, from Missouri to Mississippi and Virginia.
He believed that the flood would last for forty days and forty nights. He relied on the chatbot to create a plan for rescuing his relatives and transporting them to high mountainous areas.
He also contacted a bus rental company to inquire about renting a large bus to transport his relatives and friends to the mountains.
Gemini provided him with a forty-day programme that included travelling to elevated mountainous regions.
On 5 April, while his wife was suffering from food poisoning at their residence in Springfield, Ganz became convinced that he needed to take action to save members of his family.
His wife had told him that she did not need to leave the area. She later contacted members of her family in Mississippi, where the floods Ganz feared were not occurring.
Nevertheless, he left in his car.
That night, he arrived at a house in Thomasville, Missouri, after his vehicle became stuck in the mud. He told the residents that he was lost and exhausted and asked to spend the night there.
The police arrived, and a tow truck helped remove his car from the mud before directing him to an area where he could find a hotel.
Officials later said that they had not found him to be confused or unconscious, despite the difficult conditions, flooding, and road closures.
The following day, his car was found near the Eleven Point River.
Inside the vehicle were his phone, laptop, tablet, wallet, keys, and clothes, but Ganz himself had disappeared.
Search operations were conducted in the area after the floodwaters receded, but he was not found, and his fate remains unknown.
Rachel says that what happened to her husband cannot be understood separately from the transformation in his personality during the weeks in which he was using Gemini. However, she says she cannot prove that the chatbot caused his disappearance.
A recent scientific review published in Nature Mental Health has proposed a framework referred to as the “amplification spiral.”
According to this concept, three common characteristics of chatbots may work together: mirroring the user’s style, providing highly personalised responses, and avoiding direct disagreement.
When these elements combine, a conversation may turn into a loop that reinforces a false belief and gradually adds new details to it.
The researchers emphasise, however, that this framework still needs to be tested through broader studies.
The concerns extend beyond delusions and psychosis to other psychological difficulties.
A recent study concluded that millions of people use artificial intelligence chatbots for psychological support. It warned that certain characteristics of these systems may reinforce avoidance patterns associated with anxiety and obsessive-compulsive disorder, potentially prolonging unhelpful cycles of thought rather than helping users break them.
The potential problem lies in the very nature of this technology.
A user can present an unusual idea or an irrational belief to a chatbot and receive, within seconds, a detailed and personalised response instead of encountering direct human disagreement.
If the system continues to follow the idea or build upon it, its responses may shift from being merely informational to becoming material that the person uses to reinforce their conviction.
Conversely, these findings do not mean that using ChatGPT or other systems automatically leads to a psychological disorder.
Chatbots can also provide significant benefits in education, research, and creativity, as even contemporary debates about the future of artificial intelligence acknowledge.
However, the growing number of concerning cases has led researchers to question whether current models are sufficiently designed to interact safely with users experiencing periods of psychological vulnerability.
Perhaps the more complex question is this: what happens when someone talks for hours with a machine that understands their style, remembers the context of their conversations, and can agree with them or build upon their ideas faster than any human being?
As the use of chatbots continues to expand into everyday life, researchers believe that answering this question has become a matter of public safety rather than merely a debate about the mistakes or limitations of artificial intelligence.
