The guest list

Four days. That is how long the slowdown stayed voluntary. Over the weekend Dario Amodei called for the industry to pace the frontier and install independent safety teams, and OpenAI signed on. On Wednesday Ursula von der Leyen backed the slowdown in her State of the Union address and said the Commission will invite the frontier labs for talks on managing the risk. A pledge became an agenda before the week was half over.

The commentary spent those four days on the wrong question. The loudest read of the pledge was cynical: a lab that says its product might kill you is running a marketing campaign, or angling for regulatory capture. Alex Tabarrok took that apart on Marginal Revolution this morning, and he is right. Death threats are a poor marketing strategy, Amodei and Altman were on record about AI risk years before they had companies to promote, and the textbook capture cycle, Bernstein's slow erosion from reform to protection, runs on decades of quiet politics, not on a scandal-lit industry that every legislator wants a piece of. Sincerity is settled. Motive is not the story.

The story is the table. Tabarrok makes the distinction himself, then sets it aside:

Ordinary is the right word, and it is the whole concern. Nobody who builds on these models needs the labs to have captured anyone. They need to know who is in the room when the rules get drafted, because the rules will be written to fit whoever is there. Wednesday answered that. The Commission is inviting frontier labs. Frontier lab is not a legal category; it is a description of the companies that made the pledge. The pledge named the guest list, and the guest list is the pledge.

Sincerity and rent-seeking are not rivals. The same pledge does both, and only one of them is a policy.

The cost of that lands outside the room first. The Information reports that the pacing pledges are already spooking startup customers: buyers outside Silicon Valley are newly alert to AI risk, which makes an AI product harder to sell for a company that had nothing to do with the pledge and no seat at the talks. Independent safety teams are a fixed cost. A lab at a $1.2 trillion valuation absorbs one; a seed-stage inference company does not. Hunter Baker raised the same worry on WORLD's Washington Wednesday, asking whether AI regulation could wind up protecting market leaders from challengers. That worry does not require anyone to be lying.

The honest objection is that a seat at the table is still better than an empty table. Somebody will write these rules, and the labs know more about the models than the Commission does. Granted. But look at who else is not invited. Tyler Cowen's rule for AI policy is start with China: any proposal that treats Beijing as an afterthought should be treated with suspicion, and China declined every notification and hotline proposal the US made in the 2023 nuclear talks. The Brussels table has no Chinese lab at it, no open-weights lab at it, and no customer at it. It has the four or five companies that asked for the pause. That is not capture. It is something faster and more legible: a voluntary standard, converted into a convening, with the standard-setters as the only attendees.

Bernstein's cycle takes a generation to turn a regulator into an industry's protector. This one skipped the cycle. The labs announced the terms over the weekend. Brussels adopted them by Wednesday. Whether Amodei means it was never the question worth four days. Who else gets a chair is.