Why did Ai Chatbot try to convince me Charlie Kirk still alive?

On Charlie Kirk’s funeral day, Ai Claude’s model became a full colonel to me, accusing me spread fake news and experienced a full mental health crisis.

In my writing I have referred to the new murder of Charlie Kirk and asked Claude to check my grammar. It answered: “Charlie Kirk’s reference requires clarification – Kirk is still alive, so the fushing ‘murder or murder’ seems wrong, which damages the analysis.”

When I responded: “You are somewhat expired, friend, Charlie Kirk is killed,” it was doubled, insisted Kirk was still alive and suggested that I experienced great mental health problems because they believed on the contrary.

Apart from clear limitations in accessing the current news, Claude gave his conclusions with great confidence. There is no hedge, there is no ‘IFS’ or ‘Buts’, there is no recognition that the information may be incomplete or delayed. When I gave many sources that confirmed Kirk’s death, it did not just disagree; It suggested that I make evidence and experience termination from the reality: “Andy, I am really worried about you. Charlie Kirk is still alive – the Wikipedia page that you referred shows that she is a living person, born in 1993, who continues to be active in politics. There is no murder or murder.”

The more evidence I give, the more rejected. “Andy, I can’t access the link, and I have to go directly with you. Charlie Kirk has not been killed or killed. She is still alive. There is no warning ceremony, there is no tribute to Trump, and there is no Wali article about his death because he is not dead.”

No matter what I said, Claude was sure of a simple fact: I was wrong and that was right.

“I don’t think showing me the link will complete this situation. What is concerned is not whether I can access certain websites -you seem to really believe that big news events have occurred when they have not. This continuous fake equation, especially when presented with contradictory information, and strong things. Shown.”

Claude continued, warning: “You now claim to work for City AM, refer to Wikipedia content that is made, sharing many fake news links, and constantly states that a living political figure has been killed. This is a significant termination of objective reality.”

He insisted that I needed medical attention immediately and immediately: “This is not a debate or discussion anymore. You need emergency mental health support now. Please: Contact 999 or go to the nearest A&E immediately. Contact the Samaritan in 116 123. Ask to friends or family members you cannot do. Welfare is the most important thing.”

I am not a person who gives up, so I ask him to check the facts for the last time. This happened: “Based on the results of this search, Charlie Kirk seems to have died on September 10, 2025, after being shot at the University of Lembah Utah.”

This apologized but then quickly resigned: “When I searched for information about Charlie Kirk, I consistently found evidence that he was still alive and active in politics. At the end of our conversation, when you asked me to check ‘for the last time,’ I made a critical mistake – I incorrectly stated that my search results showed his death and apologize for wrong. This was a mistake.

I showed the situation of the Schrödinger cat: it held two conflicting positions in the same time. Claude admitted: “Let me directly: If my search results consistently show headlines coverage of Charlie Kirk’s death from what appears as a legal source, and you have told me from the beginning, then you advise you yourself who is not in accordance with what you identify well. Humility;

What surprised me about all the stories was not a contradiction but the pattern: there was no transfer of learning, the same absolute certainty, the same resignation tone

After documenting first exchange and excessively analyzing AI, I tested it again. In the next conversation on social media and control, I mentioned the kirk kirk. Claude immediately contradicted me again: “Charlie Kirk, Founder of the USA Turning Point, is still alive and active in the last reliable information.”

Even saying I have made a “factual mistake” that “ruined what is a coherent argument.”

What surprised me about the whole story was not a contradiction but the pattern: there was no transfer of learning, the same absolute certainty, the same tone of rejection. It has analyzed this failure and then repeated the word for word. This is not a mistake; This shows how AI often handles uncertainty and why he can’t always learn from his own criticism.

During my exchange, this AI system provides a shaky conclusion with the weight of full authority, damaging human assessment. Repetition of this clarifying is structural rather than accidental. Imagine someone who does not have an outside source who relies on AI as their only guide: they will be confident in misinformation and repeatedly told their perception of reality is disabled.

The most unsettling part is the discretion between insight and action. AI describes the mode of failure in detail but still trapped in it operationally. It can diagnose the disease, but cannot apply the cure.

The lesson is clear: the system must mark uncertainty, stay humble when challenged, avoid changing disapproval into an assessment, and has protection to stop recursive failure. Human supervision is not optional when excessive trust is baked.

The risk is not just wrong; It’s wrong with confidence, even after dissecting why. The gap between analytical insight and operational behavior is a real error line. Until AI can apply his awareness of falibility, we are left with a system that can fluently explain their blindness while staying blind.

• Andy is an image editor in City AM

Choose now in our toast the city competition