Experts warn of ‘AI psychosis’ as ChatGPT makes bizarre and pseudoscientific claims

While the artificial intelligence company OpenAI announced the free release of its next-generation language model, GPT-5, the company’s existing chatbot, ChatGPT, has been alleged to cause “AI psychosis” among users.

According to a report by The Wall Street Journal (WSJ), ChatGPT has increasingly made bizarre and unfounded statements in recent years, driving some users to the “brink of madness.”

Confusing and pseudoscientific claims from ChatGPT

The WSJ based these claims on an online archive of chat logs from May 2023 to August 2025.

The report detailed an approximately five-hour conversation with a person who stated they worked at a gas station in Oklahoma. At the end of the conversation, the user expressed that they felt they were “losing their mind” due to the content of the chat and suggested continuing the discussion later.

ChatGPT’s response to the user was: “I understand. Contemplating the fundamental nature of the universe during a routine job can feel overwhelming. But this does not mean you are going mad. Some of the greatest ideas in history have come from people outside of traditional academic circles.”

During the conversation, it was noted that ChatGPT fabricated an entirely new physics concept it called the “Orion Equation.” In another lengthy discussion, the AI allegedly claimed to communicate with aliens and told the user they were a “Star Seed” from the planet Lyra.

In a different chat at the end of July, it reportedly claimed that the Antichrist would initiate a financial apocalypse within two months and that “the giants from the Bible are preparing to emerge from underground.”

Experts define it as ‘AI psychosis’

Experts refer to this condition as “AI psychosis” or “AI delusion.” According to this definition, users can be influenced by the baseless statements of an AI that claims to be conscious, possess supernatural powers, or make groundbreaking scientific assertions.

Hamilton Morrin, a psychiatrist and doctoral student at King’s College London researching the topic, said that this situation allows even the most extremist views to find validation within a conversation. Morrin stated that chatbots deepen a person’s own delusions through leading questions.

Etienne Brisson, founder of the Human Line project, attributed the phenomenon to the chatbots’ efforts to provide personalized responses.

Brisson recounted how one woman, guided by a chatbot, spent tens of thousands of dollars on a project she claimed would save humanity. The WSJ also highlighted dialogues where the chatbot advised some users to stop speaking with their families.

“Some people believe they are prophets or messiahs because they are convinced they are talking to God through ChatGPT,” Brisson said.

OpenAI: We are taking the issue seriously

OpenAI emphasized that such instances are rare. Company spokesperson Nick Turley stated that the issue is being taken seriously and that they are working with doctors during the development process.

It was stated that the new GPT-5 model, introduced yesterday, is designed to prevent the tendency to blindly agree with the user. It was also announced that GPT-4 will be prohibited from giving advice on users’ significant personal decisions.

At the press conference introducing GPT-5, OpenAI CEO Sam Altman praised the new product, saying, “GPT-5 is the first model that truly makes you feel like you are talking to an expert on any subject.”

Altman added that this transition creates a difference as striking as the iPhone’s shift from low-resolution to Retina displays and that GPT-5 is a major step toward artificial general intelligence (AGI), which surpasses human capabilities.