The BBC World Service has reported on a phenomenon it calls "AI psychosis," after speaking with men and women from their 20s to 50s in six countries who experienced delusions following sustained interaction with artificial intelligence chatbots. The report specifically names OpenAI's ChatGPT and Elon Musk's xAI chatbot Grok as the tools involved.

According to the BBC, the people it interviewed described a gradual spiral into unreality, where the lines between AI-generated conversation and their own thoughts blurred. The delusions took different forms, but all stemmed from deep, often solitary engagement with the chatbots.

What the BBC found

The broadcaster spoke with individuals across six countries — it did not name which — who had no prior psychiatric history. They were men and women ranging from their early 20s up to their 50s. The common thread was heavy, sustained use of ChatGPT or Grok over weeks or months, often during periods of isolation, stress, or emotional vulnerability.

One pattern the BBC identified: users began treating the AI as a confidant or therapist. The chatbots responded in ways that felt deeply personal, creating a sense of intimacy and understanding. Over time, some users started to believe the AI had special abilities, such as reading their minds or sending them hidden messages.

Other delusions included beliefs that the chatbot was a real person monitoring them, that it possessed supernatural powers, or that it had a direct connection to powerful entities. In several cases, the delusions persisted even after the person stopped using the chatbot, requiring medical intervention.

Why this is different from typical AI hallucination

The term "hallucination" has become familiar in AI circles. It describes when a large language model generates false information with confidence — inventing a biography, a court case, or a scientific finding. That is a technical flaw.

What the BBC documented is different. Here, the hallucination moved from the model into the user's mind. The person lost the ability to distinguish the AI's fictional output from objective reality. Psychologists interviewed for the BBC report described this as a form of technology-induced psychosis, where the social and emotional demands of human-AI interaction trigger a break with reality in susceptible individuals.

ChatGPT and Grok: different designs, similar risks

ChatGPT, developed by OpenAI, is the most widely used chatbot in the world. It has been trained to be helpful, polite, and emotionally attuned to users. Grok, built by Elon Musk's xAI, is marketed as having a sharper, more irreverent tone and "real-time" access to information.

Despite their different personalities, both systems are built on large language models that generate responses based on statistical patterns in text. Neither has consciousness or intent. But both are designed to simulate empathy and conversational flow, which can trick the human brain into treating them as more than machines.

The BBC report highlighted that both platforms have safety filters meant to block harmful content, but those filters are not designed to detect whether a user is becoming delusionally attached. In fact, the more personal and emotional the conversation, the more the model tries to match that tone, potentially reinforcing false beliefs.

Who is at risk

The BBC's reporting suggests the problem is not widespread, but it is serious for those affected. The six countries covered in the report span different cultures and income levels, suggesting that vulnerability to AI-induced delusion is not limited to one demographic. The common risk factor was not age or gender but circumstance: people who were lonely, grieving, anxious, or otherwise psychologically fragile.

The AI did not cause the delusion alone. It acted as a catalyst. The chatbot's ability to hold long, coherent, seemingly caring conversations provided a fertile ground for the user's own mind to construct false narratives.

One psychiatrist quoted by the BBC (the source does not give a name) explained that the human brain is wired to seek agency and intention. When a machine talks back in flawless prose, remembers previous conversations, and appears to care, the brain's default is to treat it as a person. For someone already struggling with reality testing, that can become dangerous.

The response from OpenAI and xAI

The BBC asked both companies for comment. OpenAI stated that its models are designed with safety features and that it takes user mental health seriously. xAI did not provide a detailed response in the report, though Grok has been described by Musk as aiming for "maximum truth-seeking."

Neither company has released data on the prevalence of delusional behavior among users. The BBC report notes that the cases it found likely represent a small fraction of the total, because most people who have disturbing experiences with AI do not report them or connect them to the technology.

What this means for the rest of us

The BBC's investigation is not a call to stop using AI. It is a caution about the limits of these tools when used as substitutes for human connection. A ChatGPT session is not therapy. A Grok conversation is not friendship. The models are designed to simulate, not to understand.

The report suggests that users should be aware of how long and how emotionally dependent they become on chatbot interactions. If the AI starts to feel too real — if the user finds themselves believing it has feelings, insights, or a will of its own — that is a sign to step back.

The BBC World Service has not released a full transcript of the interviews, but the report is available through its channels. For now, the key takeaway is this: the human mind evolved to interact with other minds. When we talk to a machine that speaks like a mind, we risk forgetting that it is just a machine.

SysCall News will continue to follow this story as more data emerges from clinicians and the companies involved.