A machine can now write the exam essay. The system's answer was to march children back into a silent hall with a pen.

When something can suddenly be done by a machine for free, you learn what a human institution is really for by watching how it responds. The English exam system has just responded, and the answer repays a close read.

The threat is simple. A free tool can now produce a competent essay, a passable piece of coursework, a decent answer to most of what a GCSE asks, in seconds. That breaks the thing an exam is for. A qualification is a signal. It tells an employer or a university that a young person can really do the thing. When the machine does the work and the grade lands anyway, the signal stops meaning anything, for everyone, including the students who did their own work.

So the system moved to protect the signal. The exam boards, through the Joint Council for Qualifications, tightened the rules: pass off AI work as your own and it is malpractice, with penalties up to disqualification. Ofqual's chief regulator wrote to the boards about AI in coursework. And the enforcement is real, not a leaflet. In 2025 there were over eleven hundred cases where a student lost a whole GCSE or A-level to malpractice, and nearly two thousand more where marks were docked. The practical fix everyone is reaching for is the same one: move the work back into the room. Do it under supervision, in exam conditions, on paper, phones left at the door.

Look hard at what that means. Faced with a machine that is good at recalling facts and writing them up, the system's instinct is to build a space the machine cannot enter, and then to test, inside that space, the exact skill the machine has made cheap. It is not asking whether recalling and writing up is still the right thing to measure. It is asking how to keep measuring it cleanly. The reflex is to defend the exam, not to question it.

You see the same reflex in the year's big curriculum review, which trims the exam load but keeps the model: knowledge-rich, tested in the hall. The whole apparatus is tuning itself to preserve its own measurement. That is what systems do when they are threatened. They get better at the thing they measure. They do not stop to ask whether it is the right thing, because that question sits above the machinery, not inside it.

There is a fairness cost too, quiet but real. The detection software schools are buying to catch AI is unreliable, and one study found it wrongly flagged the writing of people who learned English as a second language as machine-made most of the time. So the crackdown lands hardest, by accident, on some of the students least able to argue back.

None of the people running this are foolish. Given the tools they hold and the job they are set, the retreat to the silent hall is the rational move. But it is a tell. It says the system was built to certify one specific human ability, the reproduction of knowledge under timed conditions, and that ability is the one the machine has just made common. You cannot fix that with a better invigilator. It is not, underneath, a cheating problem. It is a what-are-we-measuring problem, wearing a cheating problem's clothes.