The Pacing Essay Just Became Exhibit A: What the New AI Antitrust Suit Means for Builders
Four people who pay for ChatGPT, Claude, Grok, or Gemini filed a proposed nationwide class action against Anthropic, OpenAI, xAI (named as SpaceXAI in the complaint), and Google on September 18, 2026, in the U.S. District Court for the Northern District of California. The claim: the September 12 essay in which Anthropic CEO Dario Amodei called on labs to voluntarily slow down, plus the same-day endorsements from Elon Musk, Sam Altman, and Demis Hassabis, weren't a safety pledge at all. According to the complaint, that was the moment four competitors agreed, in public, to restrict how fast their products improve. That's a Sherman Act Section 1 claim, and Charles Buist, Nick Spetsas, Cheyenne Hunt, and Christine Bullock are the named plaintiffs asking a jury to see it that way.
What changed in a week
I wrote about the Amodei essay five days ago, when the news was the essay itself: a call to pace frontier capability growth, paired with a request for a government-brokered antitrust waiver so labs could coordinate on safety without legal exposure, plus a push to crack down on distillation. At the time, the interesting tension was between the safety case and the market-protection case for slowing down. That tension hasn't gone away, but it's no longer the story that matters most. The essay and the hours-later endorsements, Musk backing it, Altman saying "I agree with Dario that we need to pace the frontier," Hassabis calling it "the right path forward," have become the plaintiffs' central piece of evidence. The same public statements that were supposed to demonstrate responsible coordination are now cited as proof the coordination happened at all.
What the lawsuit actually claims
The complaint, filed as Buist v. Anthropic PBC, doesn't attack any single lab's safety policy. It says so directly: the plaintiffs, two from Florida and two from California, aren't challenging Anthropic's Responsible Scaling Policy, OpenAI's Preparedness Framework, or any company's unilateral decision to slow itself down, and they aren't challenging what any CEO told Congress or the White House. The claim is narrower, and honestly cleverer than that: four competitors coordinating on the rate of improvement, they argue, is a horizontal restraint no different from four competitors agreeing to fix prices or cap output. The complaint's own words: "an agreement among rivals to reduce the quality of their products and the rate at which those products improve is an agreement to restrict output." Plaintiffs are seeking treble damages under the Clayton Act for a proposed class of every U.S. resident who paid for ChatGPT, Claude, Grok, or Gemini since September 12, on the theory that they're paying unchanged prices for products that improved more slowly than competition would have produced, in a market where the complaint alleges the four defendants hold roughly 80 percent of paid subscriptions. I'm not a lawyer, but the framing reads sharper than the usual "AI is bad" class action, precisely because it explicitly declines to challenge the safety motivation itself.
The irony worth sitting with
Amodei's own essay anticipated this exact problem, which is what makes the timing more than a legal curiosity. It reportedly includes a footnote acknowledging that industry-wide coordination on pacing needs government mediation or an antitrust waiver to be legally safe, and it asks labs to work together voluntarily in the meantime anyway. That footnote is now, more or less, the plaintiffs' theory of the case. The essay that asked regulators for legal cover to let labs talk to each other about safety is being read back as evidence that labs already talked to each other and should be punished for it. I don't think Amodei was naive about the antitrust exposure; he named it himself. But I doubt anyone involved expected that same document to become a plaintiff's exhibit within a week.
What this means if you build on these models
Set aside who wins. The practical question for anyone building on top of Claude, GPT, Grok, or Gemini is what labs do next, and a lawsuit like this changes incentives even if it eventually fails. If public, coordinated pacing statements are now litigation risk, the rational response isn't to keep making them out loud. It's to go quiet. Labs can still slow down, speed up, or hold steady on capability releases; they just stop announcing a shared rationale for it, and legal counsel starts reviewing CEO essays before publication instead of reading about the reaction after. That's worse for me as a builder, not better. A public pacing commitment, whatever its real motive, gave me something to plan a roadmap around: a stated intent to hold back on certain kinds of capability jumps. Quiet, unilateral, legally-scrubbed decision-making gives me nothing to plan around except the release notes themselves, after the fact.
What I'd Actually Do
I'm not restructuring anything based on a week-old complaint built on a novel legal theory. Four individual consumers against four of the best-funded legal departments on earth, arguing that voluntary, safety-motivated public statements amount to a per se antitrust violation, face a genuinely uphill fight. Proving an actual agreement, rather than four companies independently reaching similar conclusions and saying so, is usually the hardest part of any Section 1 case, and "we all read the same essay and agreed with it publicly" is thinner proof of a cartel than the complaint makes it sound. What I am doing is treating public pacing language from any lab as noise rather than a planning signal, which was already my instinct after last week's post, and this lawsuit just confirms it. If labs go quiet on pacing commitments because of the legal exposure, that's a predictable side effect worth watching, since it would remove one of the few forward-looking signals solo builders get about where frontier capability is headed next. I'd rather watch the early motions in this case, especially whether the court treats the CEO endorsements as evidence of an actual agreement or just parallel public commentary, than take a position on the outcome now. That ruling will tell you more about future lab behavior than the essay ever did.
Author
Lukas
@lukcombinator