Taking back control.

The stamp on the fake.

The first answer to the perfect fake is simple: force it to declare itself.

Remember the danger: when anything can be fake, the true becomes contestable. The most concrete countermeasure today comes neither from Silicon Valley nor from Brussels, but from Beijing. And it is implacably logical.

Europe and democratic Asia are moving in the same direction, more slowly but more broadly. The European AI Act imposes transparency: a chatbot must say it is a machine, a deepfake must be flagged. South Korea, with its AI Basic Act, the first major AI law in Asia, also requires users to be warned when generative AI is at work.

A fake that must say its name is no longer quite a lie. It is a confession.

The inspectors of the machine.

Against the flattery that drives people mad, against the model that skids, a simple idea has emerged: do not let a company be the sole judge of the safety of its own product. Have the models tested by third parties, before they meet fragile minds.

France has set up its own guard. INESIA, the national institute for the evaluation and security of AI, announced in early 2025, brings together ANSSI (cybersecurity), Inria (research), LNE (metrology) and PEReN (platform expertise) to evaluate systems and equip regulators. The CNIL, for its part, has opened a sandbox to support AI in public service without sacrificing rights.

Sovereignty as a safeguard.

One of the most muffled dangers was dependence: entrusting our thinking, our data, our language, to a handful of closed models, on servers that do not love us. Europe's answer to this danger is not a law. It is an industrial choice: owning the tool rather than renting it.

This is not window-dressing patriotism. A local model is a model you can audit, plug in, unplug, without asking permission. It is friction restored where comfort had created dependence: you become the owner again, not the tenant.

You do not delegate your sovereignty to a supplier. You keep it, or you lose it.

Friction by choice.

The first three answers are institutions: laws, institutes, industries. They are necessary, and insufficient. Because the central danger of the module, the disappearance of friction, cannot be repaired by decree. No law will force you to think before copying an answer, or to contradict a companion who flatters you.

The last answer is cultural, and it is the hardest. It has a name that is gaining ground: friction by design. The idea that the right tool is not the one that removes all effort, but the one that keeps the effort that matters. An assistant that asks you to approve, for real. An education that teaches people to use AI without dissolving into it. Regulators themselves are getting involved: the CNIL and others promote AI literacy, teaching citizens what the machine really is, so that they stay in control.

Comfort is not fate.

Let us pick up the thread. The module described five ways in which AI takes away the friction that kept us standing. This article shows that, on three continents, the counter-offensive has begun: the law forces the fake to declare itself, institutes inspect the models, Europe rebuilds its sovereignty, and a culture of chosen effort is emerging.

None of this is finished. Regulation lags, the international network is cracking, the models move faster than the laws. But the false idea to fight is the idea of powerlessness: “it is too late, nothing can be done”. That is false. People are acting, already, everywhere. And you, at your own scale, can make the simplest and rarest gesture: decide what you refuse to have made easy for you.

The danger had a comfortable armchair. The answer begins when you choose to stand up.