Let’s talk about the right you were given ten days ago — and almost certainly never told about.
On the second of August, the main body of the EU’s AI Act went live. The coverage was almost all about the companies: risk assessments, data governance, logging, conformity checks, the big public registry. Fair enough. That is a lot of what the law does.
But buried in the same text is the part that was meant for you.
Article 86. The right to explanation. If a high-risk AI system is part of a decision that significantly affects you — you don’t get the job, you don’t get the loan, you get flagged at a border — you can now ask the people running that system to explain the role the machine played and the main elements of the decision. Article 85 gives you the second half: the right to complain to a national authority when you believe the system broke the rules.
Ten days in, here is the part everyone is missing. Almost nobody knows these rights exist. And almost nobody has used them.
Think about the human scene, because this is where it stops being legal jargon.
A small business owner in Madrid. Three years of clean records, steady cash flow, walks into the bank for a credit line. The application goes through the machine. Forty seconds later: no. The letter says “automated decision.” In the old world, that is the end of the story. You don’t argue with a letter. You don’t even know what to argue with.
This month, there is a question that person can legally ask. Explain the decision. Not as a favour. As a right. The bank has to answer in a way a human can actually understand.
Now, I want to be straight about how fragile this is.
The right only covers high-risk systems, and most of the machines already deciding your life are grandfathered until August next year. The law demands the “main elements” of the decision — a phrase with room in it. Trade secrets can be claimed. And an explanation can be satisfied in a way that never actually explains.
A real explanation sounds like: the system weighed your payment history against your current debt, it saw a pattern it associates with defaults, here is what drove the score. Boilerplate sounds like: “Your application was assessed in accordance with our internal risk framework.”
One of those is an answer. The other is a wall with a legal coat of paint.
But here is what still matters, and it is the thing the coverage keeps missing. This is the only individual lever in the entire Act. Everything else is paperwork on the company side — written by the company, filed with the company. This is the one power handed to a person. And rights like this do not get real through press releases. They get real the way every right in Europe got real: someone uses it, and the system either answers or refuses.
The first thousand people who ask will set the precedent. If the answer is boilerplate and brush-off, the right dies quietly, and the AI Act becomes a compliance regime with a pulse — lots of documents, no dignity. If one person gets a real answer, or one regulator gets annoyed at a non-answer, the whole thing firms up fast.
So the principle underneath all of it is simple. An explanation is the minimum of dignity. You cannot appeal to a machine. You cannot argue with it, or show it your side, or ask it to look again. The right to ask why is the difference between being judged and being processed.
A system that decides about you and cannot be asked why is not a tool. It is a verdict.
If you build or deploy systems that decide about people — and if you are listening to this, there is a decent chance you do — write the explanation before someone asks for it. Not the legal one. The plain one. The company that can explain its own model in language a customer can read is the one that won’t be caught flat-footed when the first complaint lands.
And the rest of us get the cheaper job. Next time a machine says no to you — the loan, the claim, the flag — ask why. That question is legal now.
The machine decides in seconds. You get to take your time. You’ve got this.