The UK’s tax authority has decided that the best way to catch people fiddling their tax returns is to hire an artificial intelligence system built by Quantexa, a British fintech firm, for a cool £175 million. This is, of course, deeply funny — not because AI cannot spot patterns in financial data, but because HMRC is about to spend nearly £200 million on a machine to catch taxpayers breaking the rules while remaining utterly oblivious to the fact that the tax code itself is a labyrinth of loopholes designed by people with better accountants than you have.

Let’s be clear about what is actually happening here. HMRC has a problem: people are not paying what they owe, or they are claiming things they shouldn’t, or they are using structures so complicated that even the tax office cannot work out whether it is legal. The solution, apparently, is not to simplify the tax code or hire more human inspectors who understand context and nuance. Instead, it is to buy a piece of software that will look at your bank statements and flag anything that looks suspicious.

Quantexa’s system will presumably scan millions of tax returns and bank records, cross-reference them with known fraud patterns, and spit out a list of people to investigate. On paper, this makes sense. Computers are good at spotting anomalies. They do not get tired. They do not have bad days. They process data at speeds that would make a human auditor weep.

But here is where the irony gets so thick you could cut it with a tax return form: HMRC is using AI to catch people bending the rules because it lacks the resources to enforce the rules it already has. Meanwhile, the tax code is so full of officially sanctioned loopholes that a person with a decent accountant can legally pay less tax than a self-employed plumber. Carried interest, pension contributions, capital gains relief, corporate structures with names that sound like they were generated by a random word picker — these are all legitimate ways to reduce your tax bill, and they are written into law.

The real scandal is not that people cheat on their taxes. People have always done that, and people always will. The scandal is that HMRC is spending £175 million to catch the small-time cheaters — the ones who cannot afford a lawyer to read them the rulebook — while the system itself is engineered to reward people rich enough to hire someone who knows all the tricks.

Consider the absurdity of the scenario: Quantexa’s AI spots that a freelancer has declared £40,000 in income but their bank shows £50,000 in deposits. Red flag. Investigation launched. Meanwhile, a property developer structures their holdings through a series of trusts in a way that is perfectly legal but was only invented by accountants who charge £5,000 an hour to understand it. No flag. No investigation. The system works as intended.

There is also the minor question of whether a £175 million AI system will actually work. Government IT projects have a track record that is, shall we say, mixed. The Horizon scandal — in which the Post Office used a faulty IT system to falsely accuse hundreds of subpostmasters of theft — should be a cautionary tale here. What happens when Quantexa’s algorithm flags someone as a fraudster based on a pattern it thinks it has identified, but the pattern is actually just how self-employed people in a particular industry tend to structure their finances? Do they get investigated anyway? Do they have to prove themselves innocent? Do they have to hire a lawyer to explain to a machine why they are not guilty?

The other question nobody seems to be asking is this: if HMRC cannot afford enough human staff to catch tax fraud, how can it afford to spend £175 million on software? The answer, presumably, is that the software is cheaper in the long run than hiring and training thousands of tax inspectors. Which is probably true. But it is also true that a human inspector can understand context, can ask follow-up questions, can tell the difference between fraud and just the way things work in that particular sector. A machine can spot patterns. That is not nothing. But it is also not a substitute for actually having the resources to do the job properly.

So here we are. HMRC, unable to fund basic enforcement of existing tax law, is spending nearly a quarter of a billion pounds on artificial intelligence to catch people who are not clever enough or rich enough to hire an accountant. Meanwhile, the people who are clever enough and rich enough will continue to use structures so Byzantine that even the tax authority’s own software probably cannot understand them.

The real question is not whether the AI will work. The real question is why we are pretending that this solves anything. A machine that catches tax fraud is a nice addition to the toolkit. But a machine that catches tax fraud while the tax code remains a playground for people with expensive lawyers is not a solution. It is just security theater with a neural network.

If HMRC really wanted to fix tax compliance, it would simplify the rules so much that you would not need a software system to spot fraud — because there would be nothing to hide. But that would require political will, and it would upset a lot of people with good accountants. So instead, we get £175 million worth of AI to catch the people who cannot afford not to be honest.

Trust us, the government says. We have a machine now.