Every September back-to-school season looks like the one before: the same ceremonies, the same photos in front of the gate, the same anxiety about timetables. This year, however, something has shifted without making a sound. Generative artificial intelligence models, now free, fast, and available on any phone, are as commonplace in a high schooler's backpack as a calculator was twenty years ago. Except a calculator doesn't write a philosophy essay for you.

The institution, meanwhile, moves at an administrative pace: usage charters are still vague, guidelines vary from one institution to another, and teachers are left to their own devices to detect what they can no longer truly detect. The real issue, however, is not cheating, an old subject that will always find its ways around. The real issue is what happens between the statement of an exercise and its correction: learning itself, that friction time where the student searches, makes mistakes, starts again, and finally understands.

What Friction Produced

This friction has always had a bad reputation. It was long presented as a waste of time to be eliminated, an inefficiency to be corrected. Yet, it is this friction that builds the memory of overcome difficulty, the one that allows, years later, to recognize a similar problem and know how to approach it. By eliminating this step, the tool doesn't save the student time: it deprives them of the experience that makes the next time easier.

Delegating this friction to an instant-response tool means obtaining the result without going through the process that builds competence. An assignment submitted on time is no longer proof of work done. And a student who has never had to struggle with a problem has never had the chance to learn to stick with it. The paradox is cruel: the more powerful the tool, the less it leaves a trace in the real skills of its user.

Teachers Alone on the Front Line

Teachers find themselves having to arbitrate alone, case by case, with detection methods that are already outdated as soon as they are implemented. Some give up grading certain types of homework, unable to guarantee their authenticity. Others return to in-class, supervised paper tests — a step back that has the merit of simplicity but teaches nothing about how to intelligently integrate these tools rather than fleeing from them.

This is not about banning these tools from classrooms — the battle would be lost in advance and would deprive students of a skill that will serve them throughout their professional lives. It is about clearly naming what must remain non-negotiable human effort, and protecting it as we protect an exam: with clear, shared rules, and above all, explained to the students themselves.

This requires work that neither an isolated institution nor a single teacher can undertake: distinguishing tasks where AI assistance enriches learning — documentary research, reformulation, language correction — from those where it simply short-circuits it. Without this clarification work, carried out at the national level and not left to the good will of each institution, school will continue to pretend to grade papers, while learning will have already moved elsewhere.

A Fog That Widens Gaps

This fog primarily benefits students who already have the means to get by without external help: those who know where to stop, who use the tool to verify rather than produce, because they grew up in a family environment that already instilled that rigor. The others, left to themselves facing a tool that answers everything without ever telling them where the limit is, run the risk of widening a gap they will only realize much later, during an unsupervised exam or a first job that demands a skill they thought they had acquired.

Parents, too, find themselves helpless. Many discover the extent of the phenomenon by chance, seeing their child use a language model to "practice" before a test, without always knowing if this practice constitutes legitimate help or outright circumvention. For lack of benchmarks provided by the institution, each family improvises its own rules, with results as disparate as the institutions themselves.

Other countries have begun to make decisions, with sometimes radically different approaches: outright bans on certain tools during school hours, conversely, assumed integration into curricula from middle school, or a massive return to oral evaluation to bypass the problem rather than solve it. None of these answers are perfect, but all have the merit of existing and providing a clear direction for pedagogical teams. France, for its part, is still moving forward institution by institution, which amounts to making each teacher bear a decision that far exceeds their scope.

A year after these tools became widespread in daily school life, it is urgent for the ministry to make a decision, rather than leaving each actor — student, teacher, parent — to negotiate alone with a technology that, for its part, waits for no one to evolve. The time this clarification will take will be paid for, each additional month, by unacquired skills that will be discovered too late.

*Opinion piece. The views expressed are those of the author alone.*