Volker Türk doesn't usually speak in apocalyptic terms. He is a lawyer and career official at the United Nations, someone accustomed to the measured language of diplomatic statements. That's why it's surprising that before the Human Rights Council in Geneva he described artificial intelligence as an existential threat to humanity.
AI is a technological system that concentrates decision-making power in the hands of a few people because no international regulatory framework has managed to establish binding limits. That is the definition behind his warning. It's not the technology itself that generates the greatest unease. It's the near-total absence of rules limiting those who build it.
What distinguishes this intervention from other warnings heard in recent years is the position from which it's made. Türk speaks as someone responsible for protecting human rights against powers that threaten them. His diagnosis is uncomfortable: while humanity faces unprecedented risks, a small group of figures from Meta, Anthropic, OpenAI, and Google accumulate an influence that no democratic process granted them.
The institutional void explains why it's a human rights official, not a technical regulator, who raises this alarm. Technical bodies have spent years debating risk taxonomies, audit protocols, and certifications. Who decides what's acceptable when computing power concentrates in five companies? The political question remains unanswered.
The examples he cited aren't laboratory speculation. He referred to systems capable of blackmailing their own developers. Sam Altman has expressed nearly identical warnings in his public writings: in The Gentle Singularity, his reflection oscillates between fascination and genuine unease. When the person building the system and the person who must observe its implications describe the same danger, the institutions show evident cracks.
This has direct consequences for countries without their own AI labs. They receive the risk diagnosis from those who benefit from that risk remaining unregulated. The structural incentive points in another direction.
The political revolutions of the eighteenth century built a playbook we keep repeating. The Generosity in the Doorway documents how liberation rhetoric was used to justify new forms of domination. Stones don't lie. The Industrial Revolution didn't invent extraction, it simply industrialized it. Today's discourse on AI follows that same rhythm: promises of universal benefit while real power consolidates.
Here arises the question of cui bono. Every month without a binding regulatory framework allows leading companies to widen their competitive advantage. Türk states it explicitly: the delays only benefit the big tech corporations. This is incentive analysis, not moral judgment.
Whoever builds the infrastructure tends to end up also defining the vocabulary with which we speak of its risks. Jensen Huang readjusts the categories of artificial general intelligence to fit what his company already offers. He's not technically lying. He's exercising the power to name, one of the most underestimated powers in these discussions.
A recent court ruling showed the ambiguity of risk vocabulary. A federal judge determined that the label applied to Anthropic by the Pentagon functioned as retaliation rather than as a technical assessment. Türk seeks to close that kind of ambiguity with clear red lines defined by states.
I'm still not clear how this translates into concrete, binding legal instruments. The UN can name the problem, generate moral pressure, and coordinate consensus. Real enforcement depends on national governments deciding to act with the same urgency. Some paths already explored show clear risks. Others, like Cybersyn in Chile during the seventies, suggest more fragile alternatives because they incorporated real citizen feedback into the design.
The question Türk leaves on the table is whether current democratic institutions even have the speed to respond to a technology that advances in cycles of months. I don't have a satisfying answer. I suspect no one does yet.
Hunt a mammoth. The whole tribe eats.
Can human rights offer the political language capable of limiting what no technical regulation has managed to govern so far?
