Antitrust Suit Says AI Giants Colluded to Slow Development

What happened

Four paying subscribers to ChatGPT, Claude, Grok and Gemini filed a proposed class action on September 18 in the U.S. District Court for the Northern District of California, accusing Anthropic, OpenAI, Google and SpaceXAI — the entity through which Elon Musk controls the Grok chatbot business — of illegally agreeing to slow the pace of AI development 139. The case, Buist v. Anthropic PBC, No. 3:26-cv-10693, names Florida residents Charles Buist and Nick Spetsas and California residents Cheyenne Hunt and Christine Bullock as lead plaintiffs, represented by attorney Nick Rowley and the firm Trial Lawyers for Justice 6713.

The complaint's centerpiece is a September 12 essay by Anthropic CEO Dario Amodei, We Must Pace the Frontier, which called for industry-wide coordination to slow how fast frontier AI capabilities improve in the name of safety 6913. According to the complaint, OpenAI's Sam Altman, SpaceXAI's Elon Musk and Google DeepMind's Demis Hassabis each publicly endorsed the essay within roughly an hour of publication — Musk reportedly writing that “Dario is right,” Altman saying he agreed, and Hassabis calling it “the right path forward” 71213. Every outlet covering the filing — AP, CBS, the Los Angeles Times, Bloomberg Law, Quartz, NewsNation and regional affiliates carrying the AP wire — converges on this same sequence of events as the factual basis of the suit 3459101112141516.

The suit invokes Section 1 of the Sherman Act, arguing the four companies' public agreement amounts to an unlawful horizontal restraint on competition, pled as per se unlawful and, alternatively, under quick-look and rule-of-reason theories 6. Plaintiffs seek treble damages under the Clayton Act, an injunction against coordinated pacing agreements, declaratory judgment and a jury trial, on behalf of a proposed nationwide class of paid subscribers dating to September 12 167. The complaint also alleges the four companies control roughly 80% of the paid AI-subscription market, a pleading assertion rather than any established judicial finding 6.

The theory: safety talk as an alleged output cartel

The plaintiffs are careful to frame their claim narrowly. They say they are not challenging any company's unilateral decision to slow its own development, nor are they objecting to the companies lobbying Congress or the White House for regulation or an antitrust exemption 81216. What they contest, in the complaint's own language, is competitors using the “shortcut” of agreeing among themselves to “substitute collective restraint for individual accountability” 81216. Bloomberg Law frames this as an alleged “classic output-restricting cartel,” in which subscribers kept paying the same price for products that improved more slowly than genuine competition would have produced 7.

The complaint builds a longer timeline behind the September 12 essay: a July 2026 industry working group on AI safety standards, a public proposal from Hassabis for a FINRA-style oversight body, a July statement signed by 1,386 people tied to major labs warning of “intense competitive pressure not to unilaterally slow” development, and reporting that OpenAI had asked members of Congress whether a coordinated slowdown would trigger antitrust exposure 6. Amodei's own essay anticipated the legal risk, proposing that the U.S. government mediate or at least authorize cross-lab safety conversations through a narrow antitrust waiver — a passage the Los Angeles Times, WDAM and CASRAI all note the plaintiffs now treat as evidence the companies understood their coordination required legal cover they never obtained 11126.

Political reaction and the regulatory backdrop

Coverage of the political fallout adds context largely absent from the core wire reports. KEYC and sister outlets note that President Trump dismissed the calls for AI regulation as a “conspiracy” and said on social media that he would form an “AI task force” and appoint an “AI czar,” while Senator Josh Hawley, a Republican, separately argued there is “no world” in which he would support exempting the most powerful AI companies from antitrust law 1613. That combination — a White House skeptical of coordinated slowdown talk and a senator hostile to any antitrust carve-out — underscores why the plaintiffs' preferred remedy, public regulation rather than private agreement, faces uncertain political footing 1216.

Where the reporting agrees

Across AP, CBS, the Los Angeles Times, Bloomberg Law, Quartz, NewsNation and the CASRAI explainer, the underlying facts are remarkably consistent: the filing date (September 18, reported as “Friday” in most accounts), the court (Northern District of California), the four named defendants, the four named plaintiffs, the legal theory under Section 1 of the Sherman Act, and the causal role of Amodei's essay and the same-day endorsements from Musk, Altman and Hassabis 345910121315. Outlets also agree that the plaintiffs explicitly disclaim any challenge to unilateral safety decisions, targeting only the alleged coordination 81216. And multiple outlets independently quote the same key lines from the complaint and from lead attorney Nick Rowley, including his warning about AI safety being set by “private self-serving agreements” rather than democratic accountability 1315.

Where it doesn't

The most notable divergence is tonal rather than factual. AP-derived coverage (CBS, WBKO, WDAM, KEYC, ground.news aggregations) reports the allegations in relatively neutral, hedged language, consistently describing the coordination as something the lawsuit "claims" or "argues" 91011141516. Bloomberg Law and Quartz lean into sharper framing, with Bloomberg characterizing the alleged conduct as a “classic output-restricting cartel” and Quartz citing the plaintiffs' own “publicly announced pinky promise” language more prominently in its lead 7814. NewsNation and the Los Angeles Times foreground the governance angle — the idea that private companies may be setting safety and innovation policy without public oversight — more heavily than the straight wire accounts do 1213.

CASRAI's explainer is the outlier in scope rather than substance: it is the only source to place the lawsuit inside a broader lattice of prior industry statements — the July working group, Hassabis's FINRA-style proposal, OpenAI chief scientist Jakub Pachocki's September 6 essay, and a September 10 WIRED report that OpenAI had privately asked Congress about antitrust exposure for a coordinated slowdown 6. None of the other sources mention the Pachocki essay or the WIRED report at all, which means those specific claims rest on a single account rather than corroborated reporting. Similarly, only CASRAI details the internal structure of Amodei's essay — its three-step proposal and the footnote conceding the second step needs government mediation — though the Los Angeles Times and WDAM separately confirm the government-waiver point in less granular form 11126.

There is also a minor discrepancy in how outlets describe the timing of the endorsements: several describe them as occurring within about an hour of the essay's publication 6, while others simply say the rival executives responded “the same day” without specifying how quickly 812. This is not a contradiction so much as varying precision, and it does not affect the substance of the claim.

What the evidence supports

On the facts that matter for the antitrust claim, the wire-service consensus is solid: four companies' chief executives did publicly endorse the same call for coordinated pacing on the same day, and that endorsement is now the evidentiary spine of a federal lawsuit. What remains genuinely unresolved — and the sources are honest that it remains unresolved — is whether public agreement with a safety proposal constitutes an enforceable restraint under the Sherman Act, as opposed to parallel but independently reached policy positions. No outlet reports that a judge has ruled on the merits, and the defendants had not responded publicly as of the initial filing. The case's significance lies less in likely damages than in the precedent it could set for whether AI safety coordination is legally distinguishable from output collusion, a question the reporting surfaces but cannot yet answer.