At a G20 tech summit in Chapel Hill, North Carolina, the world's twenty biggest economies pulled off something genuinely rare: they reached unanimous agreement. On what, you ask? On the urgent, hard-fought position that they would really rather not regulate AI. All of them. Including China. Including Russia. Getting a roomful of superpowers to agree that gravity exists is one thing. Getting all twenty to agree they should not build a "DMV for AI" is another kind of magic entirely.
The fruit of this glorious meeting of minds is the Carolina Principles, a US-drafted, officially non-binding framework that essentially tells the world to stand back and let AI do its thing. Apply existing sector rules. Refuse to invent new AI-specific watchdogs. Decline to build brand-new AI courts. And whatever you do, stop treating each fresh model drop like an unprecedented planetary emergency.
It is the kind of diplomatic breakthrough that makes the historian in the back of the room weep softly and the venture capitalist in the front row check their portfolio with satisfaction. Everyone finally found something they agree on: the urgent need to not pass any new laws this quarter.
The World's Strangest Unanimous Vote
The US showed up to Chapel Hill with a simple pitch: light touch. White House tech advisor Michael Kratsios summed it up: "Policymakers do not need to approach each innovation in isolation and should not treat every emerging technology as a first-of-its-kind policy problem." In human terms: please stop panicking.
Commerce Secretary Howard Lutnick dubbed the result a "historic moment of unity." Diplomatic code for the discovery that two dozen rivals, frenemies, and sworn adversaries could finally find common ground in the form of a shared refusal to legislate.
For the uninitiated, here is what the Carolina Principles actually demand:
- No new AI-specific regulators or dedicated oversight bodies.
- Existing sector-specific rules are presumably enough.
- Regulate only "genuinely novel" risks, and only when they show up.
- Everything stays non-binding until the December review in Doral.
Every tech CEO worth their equity made the pilgrimage to reinforce the message. Elon Musk shrugged that in Europe "things are generally default illegal. It slows it down quite considerably." Mark Zuckerberg reportedly told President Trump that a national AI regulator was a flawed idea, a man clearly in a hurry. Jensen Huang urged ministers to regulate demonstrated harms instead of "theoretical" ones, then helpfully pitched AI as national infrastructure, pegging a single gigawatt at USD 50 to 60 billion. That is not a plea for oversight. That is a sales pitch.
The choir even has an Oscar-nominated voice now. George Clooney took time from his busy schedule to announce that the government should not regulate AI, joining a surprisingly crowded "hands off" ensemble that apparently has room for literally everyone.
The One Person Not Invited to the Party: The EU
While the G20 perfected the art of doing nothing, Brussels had a very different night. EU tech-sovereignty chief Henna Virkkunen doubled down on the EU AI Act and declared the bloc "ready to take all necessary steps." On September 2 the EU shipped information requests to more than 30 AI companies under Article 91 of the Act, on top of Article 50 transparency enforcement that went live August 2, plus roughly 40 new enforcement staff on the payroll.
So the grand global alignment of 2026 looks like this: the G20 agreed the best AI policy is to not really have one, while Europe is hiring additional cops to police the AI it already decided to rule. Two continents, two very different vibes.
The room was not entirely of one mind. DeepMind's Demis Hassabis floated a FINRA-style financial watchdog for AI, and the room nodded politely before continuing to pass the "do nothing" resolution. Outside the venue, protesters demanded more regulation, bless their hearts, clearly not on the email chain.
The Carolina Principles roll on to the G20 Leaders Summit in Doral this December for a formal stamp of approval. Whether "let's not" survives contact with twenty actual heads of state is anyone's guess. But one thing is guaranteed: unanimity is a lot easier to achieve when the motion on the table is "let's skip this one."
Comments