Geneva, 6 July 2026. Palexpo, not Palais des Nations — the AI for Good crowd, plus 193 flags. First time every country has had a seat at this particular table.
The occasion: the first UN Global Dialogue on AI Governance. Not a summit of the willing, like Bletchley or Seoul or Paris. All of them. For the first time.
Forty scientists opened it. The Independent International Scientific Panel on AI — economists, ethicists, machine learning researchers, one from each region, none of them speaking for a government. Their first report, months in the making, three days of plenary and sixty-odd virtual sessions behind it. The headline finding, stripped of its footnotes: capability is outrunning the capacity to measure it, let alone govern it. A handful of companies, a handful of countries, hold the compute. The United States alone: three-quarters of it. Agentic systems — the ones that act rather than answer — are, in the Panel’s words, a governance step change. Nobody has a reliable way to know when one is lying to its own evaluators.
Guterres took the evidence and made it plain. Three warnings.
Speed: the internet took fifteen years to reach a billion people. AI took two. Our institutions, he said, were built to govern machines that follow commands. Not machines that decide.
Power: the imbalance hardens the longer it sits. When it’s wired into the technology, inequality becomes part of the code.
Truth: a machine-made lie now persuades as well as the truth does. A society that can’t agree on what’s real can’t defend itself.
Then four priorities, because a warning without a plan is just weather. Safety — common baselines for frontier systems, and an AI Child Safety Pledge: prove it’s safe before a child can reach it, zero tolerance for abuse material, never leave a child in crisis talking to a machine alone. Red lines — human rights non-negotiable, a human answerable for every high-stakes decision. Capacity — half a trillion in private AI infrastructure last year, a rounding error in public investment for the rest of the world; twenty-some states already signed onto a new capacity-building network, a Global Fund for AI in the works. Transparency — data centres will drink more electricity than all but five nations by 2030, enough water for all of sub-Saharan Africa for a year, and the communities hosting them are owed the numbers.
One line stayed with me longer than the rest: we cannot vibe-code the future of humanity.
Nine years ago, in his first address to the General Assembly, Guterres used the words “artificial intelligence” and almost nobody else in the room did. This time, 193 delegations. New York, 2027, is already on the calendar.
A rule only means something once it’s been tested by the people it constrains. Geneva has written its four priorities. Whether they hold is not a question anyone in that room can answer alone.
“Vibe coding” was coined by Andrej Karpathy — an AI researcher, OpenAI co-founder, former Tesla AI lead — in a viral post on X on 2 February 2025. He described it as fully giving in to the vibes, embracing exponentials, and forgetting that the code even exists — essentially, describing what you want in plain English and letting an AI generate the code, then accepting what it produces without checking it line by line.
The term is really shorthand for two different things people conflate: Karpathy’s original, provocative sense — “forgetting that the code exists” and just trusting the output — versus a more measured sense where a developer still guides, tests and reviews, just relies on AI to write more of the actual code. The word caught on fast — it earned a slang entry in Merriam-Webster by March 2025 — and the risky version is the one that stuck in the public imagination: skip the scrutiny, if it looks like it works, ship it.
That’s exactly the sense Guterres was reaching for in his Geneva remarks — “vibe-coding” as a metaphor for how societies might be tempted to relate to AI generally, not just to code: let the machine do it, don’t look too closely, seems to work, good enough. His line — “we cannot vibe-code the truth, we cannot vibe-code the future of humanity” — borrows the term precisely to warn against that same uncritical hand-off at civilisational scale.