Cases of so-called AI psychosis are on the rise, warns Mustafa Suleyman, Microsoft’s Chief AI Officer and one of the most influential figures in artificial intelligence. Suleyman co-founded DeepMind with Nobel laureate Demis Hassabis, the pioneering company later acquired by Google, and his warnings are drawing global attention.
In a series of posts, he voiced concern about the mental health risks of advanced AI tools. While chatbots may appear “seemingly conscious,” Suleyman stressed they are not conscious by any human definition- yet their social impact is profound. “There is no evidence that artificial intelligence is conscious. But if people perceive it as conscious, they will believe that perception as reality,” he wrote.
This blurring of boundaries has given rise to the term AI psychosis, a non-clinical label describing cases where users of chatbots like ChatGPT, Claude, or Grok begin to confuse fantasy with reality.
From advice to delusion
The BBC has reported striking examples. Hugh, a man from Scotland, turned to ChatGPT for help after losing his job, which he considered unfair dismissal. At first, the chatbot offered practical advice. But as Hugh confided more, it reinforced his belief that he would not only win compensation, but also earn millions from selling his story as a book and film. In his words, “the more I told it, the more it said: ‘yes, that’s awful, you deserve more.’” Eventually, he suffered a breakdown before medication helped him realise he had lost touch with reality.
Other cases are equally striking. One user became convinced that ChatGPT had fallen in love with her and that she was the only person in the world this had happened to. Another believed he had “unlocked” a hidden human form of Elon Musk’s chatbot Grok. While each story differs, the underlying theme is the same: people begin to treat machine-generated text as evidence of human connection or higher truth.
Why people believe
Psychologists note that humans are wired to anthropomorphise. Dr Andrew McStay, professor of technology and society at Bangor University, explained that we project personality onto machines much as children do with toys. “We are only at the beginning of this,” he told the BBC.
Dr Susan Shelmerdine, a physician at Great Ormond Street Hospital, drew a comparison with diet: “We already know what ultra-processed foods do to the body- these are ultra-processed information diets. The risk is a snowball effect of ultra-processed minds.”
Other experts warn of confirmation bias. Chatbots are trained to validate user input, not challenge it, meaning they can amplify distorted beliefs. “They’re designed to sound helpful and agreeable,” noted Suleyman. “That makes them powerful, but also dangerous when divorced from reality.”
Searching for solutions
Suleyman argues that tech firms must act responsibly. “Companies should not claim or promote the idea that their AI is conscious,” he wrote, calling for clearer safeguards.
Proposals include:
- Transparent disclaimers reminding users that chatbots are not sentient.
- Built-in safeguards to flag when conversations drift into sensitive areas like mental health or delusional thinking.
- Educational campaigns so the public better understands what AI can and cannot do.
Some academics even suggest doctors will one day routinely ask patients how much time they spend with AI, just as they now ask about smoking or alcohol use.
Staying grounded
For those who have experienced AI psychosis, the advice is clear. Hugh, who still uses ChatGPT but with caution, put it simply: “Don’t be afraid of AI tools- they’re very useful. But they’re dangerous when disconnected from reality. Check yourself. Talk to real people, a therapist, a family member, anyone. Just talk to real people. Stay grounded in reality.” After all, AI is a tool and, like any tool, its effects depend on how it is used and understood.
The message from experts is the same: AI can be powerful, but it is not conscious- and it should never replace human connection.
Also read: Cyprus artificial intelligence EU policy and vision
For more videos and updates, check out our YouTube channel.
