One must possess a certain courage — or a certain blindness — to organise, in 2026, yet another global summit on “AI for social good” when a report commissioned by the United Nations itself has just admitted, in black and white, that governments no longer understand the technology they claim to regulate. Yet, this is precisely what Geneva is preparing to do this week, with its live demonstrations, start-up competitions, and roundtables on digital sovereignty. A well-oiled diplomatic ritual, while the essential action is taking place elsewhere: in laboratories, in fundraising, in the battle for talent among three or four companies that, for their part, await no summit to advance.

A staggering observation, a ritualistic response

The UN report's conclusion is unequivocal: AI is progressing faster than the capacity of scientists and governments to understand and regulate it. This is a serious, almost staggering, statement coming from an institution that weighs every word. Yet, what has been the concrete translation of this alert into the week's agenda? One more summit, one more forum, renowned speakers who will repeat, with the nuances of circumstance, what everyone already knows: more international cooperation is needed, more safeguards, more dialogue between public and private sectors. Meanwhile, the real news of the sector is playing out outside conference rooms: a brain drain at Google DeepMind to the benefit of OpenAI and Anthropic, a rivalry between laboratories taking on a geopolitical dimension, and a European regulatory calendar — the AI Act — reaching its deadline without the industry truly having had time to prepare seriously for it.

Strictly reactive governance

Let us be lucid: global AI governance today operates in a strictly reactive mode. We regulate what has already happened, we frame what is already obsolete by the time the text comes into force, and we organise summits to collectively reassure ourselves that someone, somewhere, is keeping an eye on the situation. But the reality of power over this technology is not found in Palexpo; it is found in data centres, in fundraising rounds worth hundreds of billions of dollars, and in the unilateral decisions of a few laboratories that advance at their chosen pace, independently of the international diplomatic tempo.

The preventive window has closed

There is something almost touching in the UN's insistence on wanting to provide “a common scientific basis” for states before the opening of international negotiations on AI governance. But a common scientific basis is useless if it arrives after economic and technological power dynamics are already solidified. The truth is that the window for truly preventive AI governance closed several years ago, when the race for models took precedence over any serious attempt at international coordination. What is now playing out in Geneva, Paris, and Seoul is less governance than a chronicle, week after week, of an industry that advances without waiting for anyone — not even the United Nations.

A few truly binding rules

The real courage, for the governments gathered this week, would not be to multiply roundtables, but to accept this asymmetry of speed and draw the consequences: to target a few binding rules, truly applicable and verifiable, rather than consensual declarations of intent that no one, in the laboratories concerned, takes the time to read.