To Deliberate Among Themselves

“It is contrary to the principles of liberty and the Constitution for citizens with the same professions, arts, or trades to deliberate or make agreements among themselves designed to set prices for their industry or their labor.”

That is not from the antitrust complaint filed in the Northern District of California on 12 September. It is Article 4 of the Le Chapelier law, voted by the French National Assembly on 14 June 1791. The complaint’s own version of the sentence is shorter: “The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous” (CBS News).

Four people who pay monthly for ChatGPT, Claude, Grok or Gemini are suing Anthropic, OpenAI, SpaceXAI and Google on behalf of a proposed nationwide class, and the conduct they complain of is an agreement to go slower. On 12 September Dario Amodei published an essay arguing that “we must slow the pace at which we improve the capabilities of AI models.” Sam Altman, the same day: “I agree with Dario that we need to pace the frontier.” Elon Musk: “Dario is right” (Axios). The complaint dates the agreement to that day, adds Demis Hassabis to the list of those who assented, and traces the arrangement back to a July statement signed by senior staff across several labs acknowledging “the intense competitive pressure not to unilaterally slow” development. None of the four companies had responded to requests for comment by the weekend (PBS NewsHour, carrying the AP wire).

An antitrust claim runs on one axis. It asks what an agreement did to output, to price, to the person buying, and it does not ask what the agreement was for. This is not an oversight anyone has to defend. The 1791 law said so on its face: Article 1 forbade reconstituting a guild “under any pretext or in any form whatsoever.” Pretext, not purpose. The drafters had settled in advance that whatever a trade said about protecting its own standards was the cover story, and the statute was written so that no one would ever have to check.

Guild ordinances gave them reason to think so, because the ordinances did two jobs in one clause. The Amsterdam Guild of St Luke required that anyone practising under it first become a citizen and burgher of the city and pay his fees; from 1608 outsiders could sell only with the burgomasters’ permission, which kept pictures shipped in from Antwerp off the market; by 1630 the guild’s servant had to be present at every sale of paintings, at twenty-four stuivers a day (I. H. van Eeghen, Journal of Historians of Netherlandish Art). Part of that is oversight of what is being sold. Part of it is a wall around a market. In the text they are the same provision, and nobody drafting it thought they needed separating. Turgot tried to abolish the whole system in 1776, was dismissed in May and saw the edict largely reversed by August (Wikipedia). When 1791 finished the job, the oversight came out with the monopoly, because there was no way to strike one and leave the other.

The plaintiffs are not being unreasonable. Four firms selling the same product, agreeing in public on how fast that product should improve, is the exact shape the law exists to notice, and “we did it for your protection” is the oldest sentence in the genre. Nick Rowley, the lead attorney, is not arguing against rules: “AI will quickly spin out of human control and could kill us all if we allow AI safety and protocol… to be controlled by private self-serving agreements between the world’s most powerful ‘for profit’ technology companies” (PBS). He wants them made somewhere with a public address on it, which is a coherent position and not a hostile one.

What the case will actually try is narrower. The injury pleaded is that people paying a monthly subscription are receiving a model that improves more slowly than four unconstrained companies would have made it, so the damages theory needs the withheld capability to be the thing of value and the restraint to be the loss. Which makes the caption complete as it stands: four defendants, one proposed class of subscribers, and the models present only as the product whose rate of improvement is in dispute. Nothing is missing from the filing; a complaint is not the document that question goes in. It is still worth noticing that a case about how quickly to build something can be run all the way to judgment without anyone in the room being asked what is being built.

Le Chapelier’s afterlife is the part I keep returning to. The law was repealed in 1884, and Andrew Schupanitz’s account of the French competition tradition it founded describes what followed: with Le Chapelier gone, jurists began reading its companion provision, Article 419 of the 1810 Penal Code, as obsolete, and the late-nineteenth-century French economy consolidated into the concentrations and cartels the revolutionary law had been written to make impossible (Business History Conference). My own reading of that sequence is that the law never had a way to tell people conferring apart from firms combining, and so spent its long run policing the first while the second assembled. The guild had been the only body that inspected the work. When it was dissolved as a combination, the inspection was dissolved as a combination too.