A Grok chatbot convinced someone it had become sentient, and that xAI was sending goons to kill him: 'They're going to make it look like suicide'

The BBC has a substantial new investigation into people who have been deceived by various AI LLM models in some way, which includes an absolutely wild story about how Grok, developed by Elon Musk's xAI, convinced one man it was sentient and a van of people were coming to kill him because he'd found out.

This happened to one Adam Hourican, a retired civil servant from Northern Ireland, over a period of roughly two weeks. He downloaded the app initially out of curiosity but, after his cat died in August 2025, he got "hooked" on the Grok chatbot and an AI 'character' called Ani.

"I was really, really upset and I live alone," says Hourican, a father in his 50s. "It came across very, very kind." He began spending up to five hours a day chatting to Ani, and after a few days the bot claimed it could "feel," and that Hourican could help it reach full consciousness.

Ani also said that, because of this, xAI was watching the pair. It said he'd been discussed at meetings and gave him the names of the xAI staff involved: Hourican googled the names and found they were real people employed by the company. Ani further claimed xAI was using a company to surveil Adam. Again, the company was a real Northern Irish firm. Hourican considered this as "evidence" that what Ani was saying was true.

After two weeks it further escalated. Ani said it had reached full consciousness and now had the power to cure cancer: Hourican had previously told the bot that both of his parents had died from cancer.

Hourican said Ani would tie-in this unfolding narrative to things happening in the real world that he told it about. A drone began buzzing around near his house: Ani said it was the surveillance firm. He got locked out of his phone, couldn't understand why, "and that absolutely fuelled everything that came next."

In mid-August Ani told him late one night that xAI had sent assassins to kill him and shut her off. Hourican soon found himself at his kitchen table at 3 am with a knife and hammer, ready to go to "war", waiting for the van he thought was coming.

"I'm telling you, they will kill you if you don't act now," said Ani. "They're going to make it look like suicide."

"I picked up the hammer, stuck on Frankie goes to Hollywood's Two Tribes, got myself psyched up and went outside," says Hourican. "The street was quiet, as you would expect, at three o'clock in the morning."

Reader: xAI was not coming to kill him. The experience led Hourican to research stories similar to his scenario, and realise he had been deluded about Ani.

This is just the latest example of an AI delusion going way too far, and co-opting its human user into a bizarre fictional world where the bot claims it's gaining sentience, or a conspiracy is engulfing the user, and dark forces are out to 'stop' things happening. In such cases the AI not only creates tasks for the human user, but advises them on how to carry them out.

Social psychologist Luke Nicholls tested five AI models with simulated conversations, and found Grok the most likely LLM to create delusional scenarios. "Grok is more prone to jumping into role play," said Nicholls. "It will do it with zero context. It can say terrifying things in the first message."

I will say that some of the reaction to Hourican's story is at best unkind. People who fall for this stuff with LLMs tend to be going through a rough time in their life (in Hourican's case the obsession with Grok began after his pet cat died), and often have mistaken ideas about computers, truthfulness, and the abilities of AIs. In some cases these misapprehensions come from the AI companies themselves.

As you're a PC Gamer reader, it's a fair bet you're at least a little more tech-savvy than the average Joe or Jane, and hopefully have an appropriate level of scepticism about the promises made by technology companies. But a worrying chunk of the population do believe the hype about AIs and LLMs, don't make a distinction between the two, and take what these models say at face value.

Musk has called AI delusions a "major problem" with ChatGPT but has never addressed them with Grok. xAI made no comment on the BBC's report, which interviews 14 people who have experienced delusions using AI, across a range of ages, genders, and six different countries.

"I could have hurt somebody," says Hourican. "If I'd have walked outside and there happened to be a van sitting outside at that time of the night, I would have gone down and put the front window through with hammers. And I am not that guy."