Consumers Sue Anthropic, OpenAI, SpaceXAI and Google Over an Alleged AI Slowdown Pact

On 18 September 2026, four people who pay for frontier AI subscriptions filed a class-action complaint in the U.S. District Court for the Northern District of California accusing Anthropic, OpenAI, SpaceXAI and Google of agreeing to slow how fast their models improve. The case is Buist v. Anthropic, PBC, No. 3:26-cv-10693; it pleads a violation of Section 1 of the Sherman Act and seeks treble damages and an injunction, and no court has found that any such agreement exists. The conduct the complaint targets began in public on 12 September, when Anthropic chief executive Dario Amodei published an essay titled “We Must Pace the Frontier.” The archive entry behind this article, and the neighbouring policy records from the same weeks, sit in the Orca AI Incident Archive, which links each entry to the reporting it came from.

What the docket actually shows

Court records give this story a spine that the news coverage does not. The complaint is 29 pages long and was docketed on 18 September 2026, the day it was filed, with Andrew T. Tutt of Trial Lawyers for Justice as the filing attorney and Nicholas C. Rowley listed as lead counsel. The docket, as mirrored by the Free Law Project’s RECAP archive, records the cause of action as “15:1 Antitrust Litigation,” notes a jury demand by the plaintiffs, and assigns the case to Chief Magistrate Judge Nathanael M. Cousins.

The plaintiffs are Charles Buist and Nick Spetsas of Florida, and Cheyenne Hunt and Christine Bullock of California, each suing individually and on behalf of a proposed nationwide class, Politico reported on 19 September. The defendants are Anthropic PBC, OpenAI OpCo LLC, SpaceXAI LLC and Google LLC. Service was quick: OpenAI was served on 23 September with an answer due 14 October, and Anthropic, Google and SpaceXAI were each served on 24 September with answers due 15 October, according to the docket.

Nothing filed through 20 October shows an answer or a motion to dismiss from any defendant. The docket instead shows OpenAI filing a stipulation on 12 October to extend its time to respond (Doc. 28), and SpaceXAI filing a motion on 13 October to sever its part of the case and transfer it to the Northern District of Texas, or in the alternative to compel arbitration (Doc. 35). Responses to that motion are due 27 October, replies 3 November, and the hearing is set for 18 November before Judge Cousins in San Jose. Anthropic, SpaceXAI and Google had each consented to proceed before the magistrate judge by 19 October.

One filing in that stretch is worth reading literally rather than as evidence: Anthropic’s corporate disclosure of 7 October lists Google LLC and Amazon Web Services, Inc. as “other affiliates” (Doc. 17). That is a standard Rule 7.1 form about corporate relationships, not an allegation in the case, and the complaint itself does not rest on it.

The 12 September exchange the complaint is built on

The complaint’s central claim is that an unlawful agreement was “proposed in public, accepted in public, and confirmed in public” (¶2). What follows is the complaint’s account. The public statements it cites were widely reported; the inference that they add up to an agreement is the plaintiffs’ alone, and no defendant has answered the complaint yet, so none of it has been contested on the record.

On the morning of 12 September 2026, according to the complaint, Amodei published “We Must Pace the Frontier,” which declared that “We must slow the pace at which we improve the capabilities of AI models,” called for “industry-wide coordination” among frontier laboratories to set “limits on the rate of unchecked AI progress,” and said coordinated slowing would let each participant reduce its pace “without sacrificing commercial advantage” (¶2).

Within about an hour, the complaint says, Elon Musk quote-posted the essay with the words “Dario is right”; Sam Altman wrote “I agree with Dario that we need to pace the frontier” and committed OpenAI to the first step of Amodei’s plan, giving independent evaluators employee-like access; and Demis Hassabis called the essay “the right path forward” while noting that implementation details required further work (¶¶3, 73). The Los Angeles Times, running an Associated Press byline on 19 September, reports that Altman, Musk and Hassabis each publicly responded in agreement the same day.

The complaint argues the groundwork was older. It cites a July 2026 statement titled “Pacing the Frontier,” signed by Amodei, Kaplan, Jakub Pachocki, Mark Chen, Shane Legg and others, which acknowledged “intense competitive pressure not to unilaterally slow” development (¶¶49–50); Hassabis’s 14 July proposal for a U.S.-led standards body modelled in part on the Financial Industry Regulatory Authority (¶53); and a working group of Anthropic, OpenAI and Google representatives that it says met regularly from July, which The Information reported on 13 September was still meeting (¶55).

From there the complaint strings together a sequence of public statements: Pachocki’s 6 September essay “An Alien Mind,” which listed “coordinating to slow down future development” as one option (¶57); a 10 September WIRED report that OpenAI had asked members of Congress whether an industry-wide slowdown could violate antitrust law (¶58); Altman’s 11 September answer to Fortune, asked whether the labs would develop a common plan — “I think that will happen” (¶59); Altman on 14 September saying AI progress “should be slower than it otherwise could be” (¶77); and OpenAI global policy chief Chris Lehane confirming on 15 September that the company had been working with Anthropic and Google DeepMind for several weeks (¶79).

The legal theory: slower improvement treated as reduced output

Section 1 of the Sherman Act, 15 U.S.C. § 1, bars contracts, combinations and conspiracies in restraint of trade. The complaint’s core move is to treat product improvement as output. “An agreement among competitors to reduce the quality of their products and the rate at which those products improve is an agreement to restrict output,” it argues, and “it does not become lawful because the products are new” (¶6). It pleads the restraint three ways: as unlawful per se (¶140), under quick-look analysis (¶141) and, alternatively, under the rule of reason (¶142).

The relevant market is defined as paid consumer subscriptions to general-purpose frontier AI assistants — the paid tiers of ChatGPT, Claude, Grok and Gemini (¶¶96–101). The complaint alleges that the four defendants collectively account for at least 80% of those subscriptions in the United States, a figure it puts forward “on information and belief” (¶102). That is a pleading, not a measurement, and no evidence for it has been tested.

The class is defined as U.S. persons who bought a paid individual subscription directly from a defendant from 12 September 2026 onward — the day of Amodei’s essay — with a separate subclass for each company (¶¶123–124). The injury theory is that subscribers pay for access to the most capable models and to continual improvement, so slower improvement is a quality-adjusted overcharge (¶¶14, 108). The complaint concedes the point that will be used against it: because the agreement is recent and development cycles run months, “its full effect on released products has not yet manifested,” and the pleaded harm rests on altered incentives and threatened injury (¶110).

What the plaintiffs want is set out in the prayer for relief: certification of a nationwide class, a declaration that the defendants violated Section 1, damages to be trebled under Section 4 of the Clayton Act, and an injunction under Section 16 barring each defendant from maintaining any horizontal agreement on the rate at which AI products are developed or released, on training-compute limits, on limits on using AI to build better AI, on coordinated release delays, on capability checkpoints operating as agreed restraints, or on exchanging competitively sensitive information to police any of it (¶150). A jury trial is demanded.

What the plaintiffs say they are not challenging

The complaint is explicit that it does not challenge any defendant’s unilateral decisions about safety, testing, environmental impacts or the pace of its own development, and it says each defendant remains free to slow itself, retain outside evaluators, ask Congress for regulation or seek an antitrust exemption (¶13). The requested injunction is drafted to exclude independent safety measures, independent slowing, lawful safety research, compliance with government requirements, petitioning and legitimate standard-setting that does not restrain independent competition (¶151).

“Plaintiffs take AI safety seriously,” the complaint states, and argues that guardrails “should not be decided by the Defendants, but by we the people, through state and federal regulation, and through the voices of juries in our communities” (¶7). The Los Angeles Times quotes the complaint arguing that antitrust law forbids the companies from taking the “shortcut” of substituting “collective restraint for individual accountability.” Lead counsel Rowley, in a statement to Politico, said the case exists so that AI safety is not left to “private self-serving agreements between the world’s most powerful ‘for profit’ technology companies,” adding that “the rule of law should be established transparently and lawfully by our government.”

Established, alleged, and still open

• Established: the complaint was filed on 18 September 2026 in the Northern District of California as a proposed class action against four named companies; it pleads Sherman Act Section 1 and seeks treble damages and an injunction; the case is before Chief Magistrate Judge Nathanael M. Cousins; SpaceXAI has moved to sever, transfer or compel arbitration, with a hearing on 18 November 2026.

• Alleged, not established: that the September statements constitute an agreement at all; that the companies coordinated through a working group from July; that they hold at least 80% of the paid market; and that subscribers have been overcharged. Every one of those propositions comes from the complaint and has yet to be tested by a defendant’s response, let alone a judge.

• Open: whether the other three defendants answer, move to dismiss or follow SpaceXAI into arbitration; whether a class is certified; and whether any claim survives to discovery.

The archive classifies this record as a policy development, at information severity, with confidence B — documented, but from a single or secondary source — and it records no real-world harm. That grading fits what the documents support: a filing, a claim and a set of public statements, with the economic injury alleged and, by the complaint’s own admission, not yet visible in shipped products.

How the fight is likely to be framed

No defendant has argued its case in court yet, so the defence is not on the record. One published analysis, by Nitesh Daryanani for the legal-research site midpage.ai on 29 September, expects the first motion to attack whether the complaint pleads an agreement in fact rather than an “agreement in principle” under Bell Atlantic Corp. v. Twombly, and whether there is any antitrust injury yet, given the concession at ¶110 that the effects have not manifested. The same analysis calls the injunctive-relief claim harder to knock out than the damages claim, and estimates Rule 12 briefing would conclude within four to six months of service. It also flagged consumer arbitration clauses as potentially decisive; SpaceXAI’s 13 October motion asks for arbitration in the alternative.

The policy backdrop is unusually loud for an antitrust filing. The American Action Forum’s Fred Ashton argued on 7 October that Congress should reject any AI antitrust exemption, saying existing tools — rule-of-reason review of competitor collaborations, standards bodies, tort liability — are sufficient. The Los Angeles Times reports that Senator Josh Hawley said at a recent hearing that there is “no world” in which he would give the largest companies an exemption to collaborate, and that President Trump has publicly rejected calls for AI regulation.

What happens next

The near-term calendar is short and specific: responses to SpaceXAI’s transfer and arbitration motion are due 27 October 2026, replies 3 November, and the hearing is set for 18 November. Beyond that, the questions are whether Anthropic, OpenAI and Google answer or move to dismiss, whether the case stays in San Francisco, and whether a nationwide class is ever certified. There is no trial date, no ruling on the merits, and no finding that any agreement existed — and nothing in the record so far changes that.