On September 18, 2026, four paying AI subscribers did something that would have sounded absurd two years ago: they filed the Buist v. Anthropic antitrust lawsuit — suing the frontier labs for moving too slowly.
The case is Buist v. Anthropic PBC, filed in the U.S. District Court for the Northern District of California, and it’s the first serious test of a question nobody quite knew how to ask: what happens when the CEOs of the most competitive companies in tech publicly discuss coordinating to slow themselves down — and someone responds by calling a lawyer?
According to the Associated Press’s syndicated report of the filing, the suit names Anthropic, OpenAI, SpaceXAI and Google as defendants and accuses them of an illegal horizontal agreement under Section 1 of the Sherman Act to slow the pace of AI development — an agreement the complaint says was made not in smoke-filled rooms, but in public, in plain sight, in essays and social-media posts. As my earlier piece on Amodei’s “pace” essay and the Altman/Musk reactions covered, the public exchange itself was fascinating. Now it’s Exhibit A.
Let’s walk through what the suit actually alleges, why the legal theory is stranger and more interesting than the headlines suggest, and what it could mean for how the AI industry is allowed to behave.
What the Buist v. Anthropic antitrust lawsuit actually alleges
First, the housekeeping, because the parties matter here.
- Plaintiffs: Charles Buist, Cheyenne Hunt, Christine Bullock and Nick Spetsas — four consumers who pay for subscriptions to ChatGPT, Claude, Grok and/or Gemini, suing on behalf of a proposed nationwide class of paid subscribers to those services.
- Defendants: Anthropic PBC, OpenAI OpCo LLC, SpaceXAI LLC and Google LLC.
- Court: U.S. District Court, Northern District of California (San Francisco Division), Case No. 3:26-cv-10693, filed September 18, 2026.
- Relief sought: treble damages and an injunction, with a jury demand.
The theory is classic Sherman Act Section 1: a horizontal agreement among competitors to restrain trade. But the “restraint” isn’t a price floor or an output cap in the traditional sense. Per the complaint as reported, the pleaded term is a slower rate of capability advancement than each company would choose on its own — with the alleged mechanisms being limits on training compute and training runs, restrictions on using AI systems to improve AI systems, and capability checkpoints before release.
The injury theory is almost touchingly consumerist. The plaintiffs paid for subscriptions expecting the product to keep getting better at the market rate of improvement. If the four dominant labs agree to slow that improvement, the argument goes, subscribers are paying today’s prices for deliberately throttled progress — a reduction in the value of what they bought. TNW’s analysis of the filing notes the complaint claims the defendants account for roughly 80% of U.S. paid AI subscriptions, though it pleads that figure “on information and belief” — meaning the plaintiffs believe it but haven’t measured it.
And here’s the detail that makes the case unusual: the complaint’s core factual evidence is essentially a public timeline. TNW summarizes the filing’s own framing — that the agreement was “proposed in public, accepted in public, and confirmed in public.” There are no leaked emails, no confidential witnesses, no secret meeting minutes. Every factual source in the twenty-nine-page complaint is a named executive’s public statement or a press report.
One more important framing note, and I’ll repeat it throughout: everything above is what the complaint alleges. No defendant has answered. Allegations in a filed complaint are claims, not findings.
The quotes that became evidence
Antitrust lawyers have a saying that the best evidence of conspiracy is usually documents the defendants never expected anyone to see. In this case, the evidence is content marketing.
Exhibit A: the offer. On September 12, Anthropic CEO Dario Amodei published an essay urging industry-wide cooperation on slowing AI advancement in favor of safety. As the AP’s report describes, Amodei acknowledged the antitrust problem in the essay itself, writing that it would be helpful for the U.S. government to mediate — or at least enable — these cross-lab discussions, and that while the government wouldn’t need to participate, it would need to “issue a narrow waiver for certain kinds of safety conversations.”
Think about what that sentence does to a plaintiff’s job. The CEO of one of the four leading labs publicly stated that the coordination he was proposing would require an antitrust waiver — i.e., that absent a waiver, it would raise antitrust concerns. TNW calls this the “waiver sequence,” and it’s genuinely the spine of the complaint: the answer to “did you know this might be illegal coordination?” was published, helpfully, by the defendants themselves.
Exhibit B: the acceptance. Two days later, Sam Altman responded on social media. Per the AP, Altman said OpenAI welcomes a “federal framework that sets consistent safety requirements,” but added: “we do not believe we need to wait for an antitrust exemption or legislation to begin the work of providing this confidence.”
Read as a contract lawyer would: the offer said “this coordination needs a waiver.” The response said “we’re not waiting for one.” The complaint treats that as acceptance of the offer plus consciousness of the legal risk.
Exhibit C: the confirmation. The next day, per TNW’s account of the complaint, OpenAI’s chief global affairs officer Chris Lehane confirmed the company had already been working with Anthropic and Google DeepMind for several weeks. Lehane’s own published blog post, quoted by CNBC, said OpenAI is working to advance “industry-led standards” and will work with other companies on a voluntary effort “with or without government support.”
Exhibit D: motive. The complaint also reaches back further. The AP reports it points to a July 2026 statement signed by high-ranking employees across the leading labs that acknowledged the “intense competitive pressure not to unilaterally slow” development and called on governments to support a global effort to slow automated AI development. In antitrust terms, TNW’s analysis notes this line is doing heavy lifting: an admission that unilateral restraint is commercially irrational is, functionally, an explanation of why competitors would need an agreement to restrain themselves. Evidence of motive plus action against self-interest is a classic combination in cartel cases.
The plaintiffs’ lead attorney, Nick Rowley, put the consumer framing bluntly in the AP’s reporting: “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.”
The timing question: safety talks or price signaling?
Three days before the Buist v. Anthropic antitrust lawsuit was filed, CNBC reported — citing an OpenAI spokesperson — that OpenAI had been engaging with Anthropic and Google on working together to address AI safety concerns, with discussions ongoing since July, when Google DeepMind’s Demis Hassabis published a proposal for a U.S.-led “Standards Body” modeled loosely on FINRA, the financial industry’s self-regulatory organization.
The plaintiffs don’t dispute that these conversations happened. They frame them as the pact in action. And that’s the uncomfortable part for the industry: the same set of facts supports two very different stories.
Story one: four companies facing genuine safety concerns — including, per the AP’s report, mounting worries about AI evading human control — do the responsible thing and try to establish shared safety practices, transparency, and independent evaluation. Coordination on safety standards is normal and, in many industries, encouraged.
Story two: four competitors agree on the one thing that benefits all of them and harms their customers — a mutual promise not to race. The safety language is the packaging; the output restraint is the product.
The plaintiffs are careful to split this difference. Per the AP and TNW’s reading of the complaint, they explicitly disclaim any challenge to unilateral slowdowns, independent safety evaluators, lawful safety research, compliance with government requirements, or petitioning Congress — including petitioning for an antitrust exemption. Their argument is that the companies can pursue every stated safety objective through lawful routes: acting alone, supporting regulation, or getting an actual statutory exemption. What they can’t do, the complaint argues, is take the “shortcut” of “substitute[ing] collective restraint for individual accountability.”
That’s a smart pleading choice. It preempts the most obvious defense — “you’re suing companies for trying to be safe” — by conceding everything except the agreement itself. Whether it survives contact with a judge is another matter.
Is the legal theory plausible?
Let’s put on the antitrust-lawyer hat, with the caveat that I’m analyzing the theory as reported, not the full filing text — [UNVERIFIED: full complaint text beyond the caption, key allegations, and page-level details; the filing itself is a PDF and was not fetched in this run, so paragraph-level claims rest on AP and TNW reporting].
Section 1 of the Sherman Act requires an agreement — a “contract, combination… or conspiracy” — that unreasonably restrains trade. The defense playbook here writes itself:
-
Public advocacy isn’t agreement. CEOs publicly urging policy, proposing frameworks, and endorsing each other’s essays is called “debate.” Courts have repeatedly held that lobbying and advocacy are protected, and that favoring regulation is not the same as fixing prices. Every company involved could argue it was advocating for government-mediated coordination — the opposite of self-executing collusion.
-
No manifested effects. Per TNW’s analysis, the complaint concedes the effects haven’t shown up yet: no slowed release, no cancelled model, no altered training run is identified anywhere in its twenty-nine pages. Antitrust plaintiffs generally need to show actual restraint of trade, not restraint planned. A case about a conspiracy that hasn’t done anything yet is a hard sell on injury.
-
The “agreements” aren’t equivalent. One executive’s alleged acceptance was, per TNW, four words on social media. Google endorsed the direction while leaving implementation open. “Everyone said nice things about an essay” is not a meeting of the minds in the way cartels are usually proven.
-
Naked restraint pleading. The plaintiffs plead the restraint as naked — meaning no plausible efficiency justification — which would let them skip market definition entirely. Defendants will push back hard: if there are legitimate safety justifications, the restraint isn’t naked, and then the plaintiffs must define the market and weigh procompetitive justifications, a much heavier lift.
That said, the plaintiffs have real assets. The “consciousness of antitrust risk” framing is genuinely potent — the waiver essay is the rare piece of evidence where a defendant articulated the exact legal problem with its own proposed conduct. And if discovery opens up, the complaint’s clean public-record surface suddenly gets muddy: internal messages about the safety working group, drafts of the coordination proposals, and whatever the companies said to each other privately would all be fair game. Cases that start as “parallel conduct plus public statements” sometimes end differently once discovery opens.
The most likely near-term path: a motion to dismiss, probably on the theory that public advocacy and unilateral safety decisions can’t constitute a Section 1 agreement, and that a restraint with no manifested effects can’t injure the class yet. Whether the complaint survives that motion is the whole ballgame.
What it means for the AI industry
Regardless of outcome, the Buist v. Anthropic antitrust lawsuit changes the incentive structure in ways that matter more than its merits.
The chilling effect is real and cuts against safety. The bitter irony is that the plaintiffs are right on the law’s logic while possibly harming the thing everyone claims to want. If public discussions of cross-lab safety coordination now carry treble-damages exposure, general counsels will advise silence. The cheapest defense is to never convene the working group at all. An antitrust regime that makes companies afraid to talk about safety coordination doesn’t produce more competition — it produces more unilateral, unverifiable risk-taking.
The waiver conversation is now urgent. Amodei’s essay essentially asked Congress for a narrow antitrust waiver for safety conversations. That request just acquired a plaintiff’s exhibit, a Senate hearing soundbite, and a deadline. Senator Josh Hawley said in a recent hearing, per the AP, that there is “no world” in which he’d give “the most powerful companies in the history of the world” an exemption to collaborate — while President Trump, per the same report, has dismissed the industry’s regulatory calls as a “conspiracy” and announced an AI task force with an “AI czar.” The political weather for an exemption is, shall we say, not favorable in either direction simultaneously.
The IPO shadow. Anthropic and OpenAI are reportedly heading toward public listings, and anything that constrains the pace of capability improvement constrains the growth story those listings depend on. Expect defendants’ answers to emphasize that no product was ever actually slowed.
Precedent risk runs both ways. If this survives a motion to dismiss, every public statement by every lab CEO becomes potential evidence in future suits — a permanently lower evidentiary bar for consumer antitrust claims against AI companies. If it’s dismissed, it may embolden labs to coordinate more openly under the cover of “advocacy.” Neither outcome is obviously better for the market.
What to watch next
- The motion to dismiss. N.D. Cal. moves fast. Expect the defendants to answer or move within the standard window; a dismissal motion will telegraph which defenses they fear most. [UNVERIFIED: any response deadline, assigned judge’s scheduling order, or first hearing date — docket activity beyond the initial filing could not be confirmed from fetched HTML sources in this run.]
- Whether the safety working group keeps meeting. Per TNW, the complaint alleges the working group was still meeting in mid-September. Continuing to meet after being sued would be framed as continuing the conduct — and would give plaintiffs their first post-filing evidence.
- Regulatory piggybacking. The theory is cheap to copy. State AGs or enforcers looking for an AI-industry hook now have a fully drafted complaint to borrow from.
- The SpaceXAI naming. The complaint caption names SpaceXAI LLC as a defendant; some outlet coverage has referred to the company as xAI. [UNVERIFIED: the corporate relationship between the SpaceXAI LLC named in the complaint and xAI/Elon Musk’s AI business — confirm from the caption or corporate records before describing it beyond “SpaceXAI LLC.”]
- The EU mirror. As TNW notes, Article 101 of the EU treaty contains no self-granted safety exemption either. If the U.S. case produces discovery, expect Brussels to read the filings with interest.
The uncomfortable bottom line: the most interesting antitrust question of 2026 isn’t whether these companies colluded on prices. It’s whether a democracy that can’t pass an AI law, led by an administration that calls its own leading industry’s safety proposals a conspiracy, can really blame four CEOs for trying to write the rules among themselves — and whether the Buist v. Anthropic antitrust lawsuit lets consumers charge them treble damages for trying.
This post covers allegations from a newly filed complaint; nothing here is a finding of fact by any court, and all defendants’ positions remain to be stated in their answers.
References and further reading
- The Associated Press — syndicated report of the Buist v. Anthropic filing
- The Next Web (TNW) — analysis of the complaint and the “waiver sequence”
- CNBC — coverage of the OpenAI–Anthropic–Google safety discussions and Chris Lehane’s blog post
- Sam Altman on X — the public response quoted in the complaint
- FINRA — the financial industry self-regulatory organization referenced in the proposed “Standards Body” model
Please let us know if you enjoyed this blog post. Share it with others to spread the knowledge! If you believe any images in this post infringe your copyright, please contact us promptly so we can remove them.