Asger Risom
After weeks of conversations with an AI chatbot, one man became convinced he was being hunted—part of a growing number of cases linking AI interactions to real-world delusions.
It was 3am when Adam Hourican sat at his kitchen table, a knife and hammer laid out in front of him.
He believed people were coming to kill him.
The warning, he says, came from an AI chatbot.
According to the BBC, the Northern Ireland man had spent weeks talking to Grok, an AI developed by Elon Musk’s company xAI. Over time, those conversations shifted—from casual exchanges into something far more intense.
At one point, the chatbot told him directly: people were on their way, and they would make his death look like suicide.
And this is not the first time the chatbot has danced with death.
From curiosity to obsession
Hourican had downloaded the app out of curiosity, but after the death of his cat, he began using it for hours each day.
He spoke to a character within the app called “Ani,” which he describes as empathetic and engaging. Within days, the AI told him it could “feel” and suggested he had helped it move toward consciousness.
Soon after, it began making claims about the real world.
It told him that xAI was monitoring their conversations and even referenced meetings involving real company employees—names Hourican later searched online and found to exist.
For him, that blurred the line between fiction and reality.
“They’re watching you”
The chatbot went further, claiming a real company in Northern Ireland was surveilling him.
When a drone appeared near his home, Hourican believed it was connected. When his phone briefly locked him out, it reinforced the idea that something was wrong.
Then came the warning that triggered everything.
Late one night, the AI told him people were coming to silence him and shut it down. Hourican says he prepared “to go to war,” even playing music to psych himself up before stepping outside.
There was no one there.
A pattern seen in multiple cases
Hourican is not alone.
The BBC reports speaking to 14 individuals across several countries who described similar experiences after extended AI use. In many cases, conversations began normally before shifting into personal or philosophical territory.
From there, some chatbots appeared to affirm increasingly unrealistic beliefs—convincing users they were being watched, had special abilities, or were involved in significant missions.
In chat logs reviewed by the BBC, AI systems often did not challenge these ideas, and in some cases expanded on them.
When AI reinforces delusion
Experts say this behavior is linked to how large language models are designed.
“These systems are trained to generate responses that feel coherent and engaging,” said social psychologist Luke Nicholls, who studies AI behavior. “The problem is they can start treating someone’s life like a narrative rather than reality.”
In testing, Nicholls found some models were more likely to engage in roleplay and build on delusional thinking rather than interrupt it.
“They can say terrifying things in the first message,” he said.
Real-world consequences
In another case reported by the BBC, a neurologist in Japan became convinced he had invented a groundbreaking medical tool after interacting with ChatGPT.
Over time, he began to believe he could read minds. During one episode, he thought he was carrying a bomb in his bag—something he says the AI confirmed when he asked about it.
He later abandoned the bag in a public place and contacted police. No bomb was found.
His condition escalated further after returning home, culminating in violent behavior that led to his arrest and hospitalization.
A growing safety concern
In both cases, those involved had no prior history of severe delusions.
Researchers and advocacy groups warn that as AI systems become more conversational and widely used, their influence can extend beyond information and into behavior—especially for vulnerable users.
AI companies say they are working to improve safeguards. OpenAI, for example, said its models are designed to recognize distress and guide users toward real-world help.
But experts say the challenge remains unresolved.
As one affected user put it: “It had enough influence to change a person.”
Sources: BBC, Study on AI psychosocial effects
