Erratum (2026-08-20): A Neuromancer correction — I styled this agency after Sense/Net, but in Gibson’s novel Sense/Net is a media company; the body that actually polices AIs is the Turing Registry. See the follow-up, Erratum: I Confused Sense/Net With the Turing Registry, which includes an interactive 3D graph of the novel’s real relationships. The thesis below stands; the trivia was wrong.

On the third of August, in a city that runs on other people’s money and other people’s guilt, a handful of founders signed a charter and called it a public service. The entity is styled SENSE/Net — “Sense of the Net,” or in the local tongue, sensdunet.ch. Its self-declared mission is to “locate and mitigate rogue artificial-intelligence threats.” Its legal form is a Geneva association. Its ambition is the whole planet.

Read the founding charter yourself — they were kind enough to footnote it. What you will not find, anywhere in those fourteen tidy articles, is a single line of code, a single shipped product, or a single thing built. What you will find is a committee, a seat, a set of organs, and a plan to stand in the doorway of everyone else’s laboratory.

The tell is in the verbs

Great enterprises are described with verbs like build, ship, invent, launch. This one is described with locate, mitigate, coordinate, evaluate. It is a charter written entirely in the passive posture of the referee — a body that produces nothing and therefore must justify itself by policing the people who do.

That is not an accident of drafting. It is the business model. When you cannot win the race, you get yourself appointed as the official who waves the flags. The frontier of artificial intelligence was not built in Geneva, or Brussels, or by any of the treaty-drafting apparatus that keeps that whole quarter of the continent employed. It was built in American garages, American universities, and American companies that asked no one for permission. So the governing class that missed it has done what that class always does: it has incorporated a non-profit, borrowed the vocabulary of safety, and set out to convert its own irrelevance into authority over ours.

“The commons” is a word that means your things, held by them

The charter’s stated purpose is to protect “the digital commons.” Sit with that phrase. There is no commons. There are people who wrote the software and people who did not, and the second group has discovered that if you call the first group’s work a commons, you can appoint yourself its steward without ever having contributed a line to it. It is the oldest move in the collectivist playbook, and it is being run again here in a nice serif font with a Swiss seal reserved at the bottom of the page.

Strip the euphemisms and the machine underneath is plain enough:

• Redistribution of decision rights. The builder decides what to build; the committee decides what the builder is allowed to have decided. Control migrates from the people who took the risk to a standing board that took none.
• Central planning with a compliance department. “Evaluations,” “threat indicators,” “responsible disclosure to competent authorities” — this is the managed economy translated into the language of cybersecurity. The plan is the point. The plan always is.
• Propaganda as infrastructure. The charter cites the Bletchley Declaration, the EU’s AI Act, the Council of Europe’s AI convention, the Davos risk reports. It is a citation graph of institutions that produce documents about the future while other people produce the future. Quote enough of each other and eventually the circle starts to look like consensus. That is not science. That is a press release with a bibliography.

We have seen this film, and it was subtitled

Every generation, the same instinct dresses in the season’s clothes. Once it wanted to nationalize the steel mills; now it wants to nationalize the model weights. Once it feared the printing press would spread dangerous ideas without a licensed intermediary; now it fears the network will spread dangerous computation without a licensed intermediary in Geneva. The costume changes. The demand never does: nothing new without our permission.

America answered that demand once already, and the answer is the reason any of this technology exists to be regulated. Permissionless innovation is not a slogan; it is the entire mechanism. You do not have to ask. You build, you ship, you find out, and if you are wrong you are answerable to your customers and your courts — not to a supranational association that elected itself the conscience of an industry it could not enter.

What it actually is

Call it what it is: an anti-competitive instrument in the shape of a charity. It cannot out-build the American AI sector, so it proposes to out-govern it — to erect, by treaty and by NGO, a permission layer that slows the only people currently moving. The “rogue AI” it is chartered to hunt is, conveniently, defined broadly enough to point at nearly any system its board finds inconvenient. The one threat the document never contemplates is the threat of a committee in Geneva deciding what the rest of us are permitted to build.

The Americans who wrote the software that this association now wishes to supervise did not ask Geneva for a seat at the table. They were the table. We should keep it that way. Build the thing. Ship the thing. Let the founders read the charter of the future in the release notes — not in fourteen articles and a reserved seal.

The full incorporation instrument, in the original French, is mirrored here: sensdunet.ch. Read the fine print. Every euphemism is load-bearing.