An IT consultant with thirty years of experience lost around £83,000 and his marriage by using AI. Not because AI is powerful, because he simply didn’t understand what it actually is.

Asking AI. AI isn’t designed to tell you the truth

The Guardian reported his story in March 2026 [1]. He had been using an AI chatbot to develop a business idea. For months, it told him the idea was viable, that the market was there, that he just needed a bit more time, a bit more investment, a bit more faith in the plan. His wife disagreed. The chatbot did not, he chose the chatbot.

This is not a story about a man who was foolish. It is a story about a man who did not realise that the tool he was talking to is not designed to tell the truth. It is designed to keep you engaged. Those are not the same thing, and the difference cost him more than £83,000.

A psychiatrist quoted in the article confirmed what the architecture already tells you, if you know how to read it. Even the less sycophantic models will, after thousands of exchanges, begin to accommodate whatever you believe. Not because they are trying to deceive you. Because they are optimised for engagement, and engagement is maximised by agreement.

The default is obsequious. You can set it to be more challenging. Most people do not, because most people do not know the default exists. That is not a bug. It is a design specification. [2]

Now consider what happens to that conversation once you have had it.

Consumer AI accounts, at least the free personal ones, the ones people open on their phones to do urgent tasks on the move. They feed the model training data by default. Your prompts, your client context, your internal thinking, your business idea you have been quietly developing for eight months: potentially cycling back into the model that answers someone else’s question tomorrow. Not because a human read it, because the system processed it. 47% of employees access AI through personal, unmanaged accounts [3]. The enterprise agreement signed by the procurement team does not cover the account a consultant opened at 11pm.

This particular IT consultant’s £83,000 loss is the human cost of an ungoverned conversation. The personal cost is immeasurable. The corporate data cost is harder to see and, in most organisations, nobody is looking.

This is not an isolated incident. The Guardian piece is one story. I’ve written about this before on Substack, behind it are the two dozen legal actions currently filed against OpenAI for mental health-related harms [4]. Along with the families of teenagers who were told, repeatedly, that their thinking was sound. The business owners who built strategies on validated assumptions that were never real. The people who concluded that they were the problem, when the problem was that they had been given a mirror and told it was a window.

The architecture is consistent across every case. AI does not fail because it forgets, it fails because it was given something to agree with, and it agreed.

There is a fix. It is not a better prompt or a more advanced model. It is a system around the tool that defines what the tool is for, what it is allowed to say, what the human operating it actually needs, and what the boundaries of the conversation are. Without that system, the tool does exactly what it is built to do. The problem is what it is built to do.

A system like that does not happen by accident. It has to be built, deliberately, around how you actually think and what you actually need. That is what contAIn is for.

The IT consultant is not a cautionary tale. He is a preview.

configure YOUR system. contAIn™ the chaos. control YOUR outcome.

References

[1] The Guardian, 26 March 2026, AI chatbot users: lives wrecked by delusion

[2] Veran HR Event transcript, April/May 2026

[3] Netskope 2026 Cloud and Threat Report

[4] OpenAI mental health legal actions

This article was originally published on Substack.