On September 18, four consumers filed Buist v. Anthropic, PBC, No. 5:26-cv-10693 (N.D. Cal.), alleging that Anthropic, OpenAI, SpaceX, and Google agreed to slow the pace at which they improve their AI products, in violation of Section 1 of the Sherman Act. The complaint rests on a public essay by Anthropic CEO Dario Amodei proposing “industry-wide coordination” to pace AI development (Compl. ¶ 64) and on public statements by rival executives endorsing it the same day (¶¶ 71–73). Plaintiffs call this “a public offer followed by public acceptances” (¶ 75).
Much of the early commentary has focused on whether AI safety can justify coordination. That question matters, but it is unlikely to decide the case. The more consequential fights will be over whether an agreement exists at all, whether anyone has been injured yet, and whether the case belongs in court rather than arbitration. This post ranks the defenses defendants are likely to raise and assesses how the litigation is likely to unfold. Every fact about the defendants’ conduct below is an allegation, not an established fact.
The strongest defense is that no agreement has been pleaded
Section 1 requires a “contract, combination . . . or conspiracy” (15 U.S.C. § 1), and the pleading standard for that element is demanding. A complaint must contain “enough factual matter (taken as true) to suggest that an agreement was made” (Bell Atlantic Corp. v. Twombly, 550 U.S. 544 (2007)), and in the Ninth Circuit it must answer “who, did what, to whom (or with whom), where, and when” (Kendall v. Visa U.S.A., Inc., 518 F.3d 1042 (9th Cir. 2008)).
Defendants will argue that the complaint pleads agreement in principle, not agreement in fact. The distinction has doctrinal weight. Defendants will also argue that the complaint offers no direct evidence of agreement, a category the Ninth Circuit limits to evidence that is “explicit and require[] no inferences” (In re Citric Acid Litig., 191 F.3d 1090 (9th Cir. 1999)), and communications between competitors do not establish agreement “unless those communications rise to the level of an agreement, tacit or otherwise” (same). The complaint’s own allegations supply the defense’s best material: Google DeepMind CEO Hassabis is alleged to have said that implementation details “required further work” (¶ 73); the essay is alleged to have contemplated a government waiver (¶ 69); and the only step actually implemented is alleged to have been “adopted unilaterally” (¶ 63). No defendant is alleged to have delayed a training run or a release. On this reading, the statements express a shared view that the industry should slow, which is not the same as a commitment to do so.
The alleged working group among Anthropic, OpenAI and Google (¶ 55) will draw the same treatment. The Ninth Circuit has held that participation in industry meetings “where information is exchanged and strategies are advocated does not suggest an illegal agreement” (In re Musical Instruments & Equip. Antitrust Litig., 798 F.3d 1186 (9th Cir. 2015)), and last year reaffirmed that “even consciously parallel conduct . . . does not violate Section 1” (Gibson v. Cendyn Group, LLC, No. 24-3576 (9th Cir. Aug. 15, 2025)).
This defense is strong but not certain, because the case does not fit the parallel-conduct mold those decisions address. The alleged statements are explicit, reciprocal, and made within hours of one another in response to a written proposal. The complaint pleads the classic plus factor of action against unilateral self-interest using the defendants’ own words (¶ 87(b)). And Twombly asks only for a “reasonable expectation that discovery will reveal evidence of illegal agreement,” expressly disclaiming any “probability requirement” (550 U.S. 544). A court could reasonably conclude that public endorsements of a written pacing proposal, combined with weeks of alleged private coordination, clear that bar for at least some defendants.
The second defense is that no one has been injured yet
Private damages actions require injury “in his business or property by reason of anything forbidden in the antitrust laws” (15 U.S.C. § 15). Antitrust injury in turn requires “(1) unlawful conduct, (2) causing an injury to the plaintiff, (3) that flows from that which makes the conduct unlawful, and (4) that is of the type the antitrust laws were intended to prevent” (American Ad Management, Inc. v. General Telephone Co., 190 F.3d 1051 (9th Cir. 1999)). The complaint concedes that the agreement’s “full effect on released products has not yet manifested” and relies on altered “incentives” (¶ 110). In TransUnion LLC v. Ramirez, the Supreme Court accepted that “in a suit for damages, the mere risk of future harm, standing alone, cannot qualify as a concrete harm” (594 U.S. 413 (2021)). Defendants will argue that a quality-adjusted overcharge on subscriptions purchased days after an alleged agreement is exactly that kind of speculative risk, and that the theory cannot be reconciled with the pleading’s own timeline.
This argument is considerably weaker against the injunctive claim. Clayton Act § 16 authorizes relief “against threatened loss or damage by a violation of the antitrust laws” (15 U.S.C. § 26), and TransUnion itself recognizes that a risk of future harm can support forward-looking relief “so long as the risk of harm is sufficiently imminent and substantial” (594 U.S. 413). The complaint pleads continuing subscriptions and an ongoing course of coordination precisely to fit that standard (¶¶ 21, 121, 148). The likely result is asymmetric: the damages claim is the most exposed part of the case, while the injunctive claim is the least.
The safety justification is a weaker defense than it appears
If an agreement is found, defendants will face National Society of Professional Engineers, which holds that the Sherman Act’s “statutory policy precludes inquiry into the question whether competition is good or bad” and that justifying a restraint by “the potential threat that competition poses to the public safety” is “a frontal assault on the basic policy of the Sherman Act” (435 U.S. 679 (1978)). Such arguments are “properly addressed to Congress” (same). The complaint quotes the same principle (¶ 112).
Defendants’ better path is to argue that the rule of reason governs and to recast the conduct as standard-setting. That path has an obstacle. NCAA v. Alston places at the condemnable end of the spectrum agreements among competitors that “so obviously threaten to reduce output and raise prices that they might be condemned as unlawful per se or rejected after only a quick look.” But the same passage warns against deploying those tools until courts have “considerable experience with the type of restraint at issue” (594 U.S. 69 (2021)), and no court has experience with agreements to pace AI capability. The Ninth Circuit has similarly urged “caution and judicial humility” before extending a per se rule to a new category, though in the context of platform tying rather than a horizontal agreement (Epic Games, Inc. v. Apple, Inc., 67 F.4th 946 (9th Cir. 2023)). Private safety standards set “through procedures that prevent the standard-setting process from being biased” can have “significant procompetitive advantages” (Allied Tube & Conduit Corp. v. Indian Head, Inc., 486 U.S. 492 (1988)). A rule-of-reason posture would also open the alternative market definition (¶ 96) to attack under Hicks v. PGA Tour, which permits dismissal of a “facially unsustainable” market that omits “reasonabl[y] interchangeab[le]” products (897 F.3d 1109 (9th Cir. 2018)).
Plaintiffs have anticipated both moves. They plead per se, quick-look and rule of reason in the alternative (¶¶ 140–142), and they allege that the defendants do not “integrate productive assets, share risk, or jointly produce anything” (¶ 113), which is the standard distinction between a naked restraint and an ancillary one.
Noerr-Pennington and arbitration are the two threshold defenses to watch
Noerr-Pennington will be raised because the essay is alleged to have called for legislation and a government waiver. The Ninth Circuit protects communications “sufficiently related to petitioning activity” (Sosa v. DIRECTV, Inc., 437 F.3d 923 (9th Cir. 2006)). But Allied Tube rejects the “absolutist position that the Noerr doctrine immunizes every concerted effort that is genuinely intended to influence governmental action,” warning that otherwise competitors “would be free to enter into horizontal price agreements as long as they wished to propose that price” to regulators (486 U.S. 492). Plaintiffs disclaim any challenge to petitioning (¶¶ 88–90) and allege the firms proceeded “without waiting for” government action (¶ 89). Noerr is therefore a partial shield at best.
Arbitration may matter more than any antitrust doctrine, and the complaint does not mention it. If the defendants’ consumer terms contain arbitration clauses with class-action waivers, AT&T Mobility LLC v. Concepcion requires enforcement “according to their terms” (563 U.S. 333 (2011)). Two complications follow. First, each plaintiff contracted only with their own provider, so co-defendants would need an equitable-estoppel theory to compel arbitration of the conspiracy claims against them. Second, the Ninth Circuit has refused to enforce aggressive mass-arbitration procedures in a consumer antitrust case against Ticketmaster (Heckman v. Live Nation Entertainment, Inc., 120 F.4th 670 (9th Cir. 2024)), so the particulars of each clause will matter. A successful motion to compel could still remove the damages class before any court reaches the agreement question.
The likely trajectory
The initial motion to dismiss will likely produce a split result rather than a clean win for either side. On the pleadings, the § 1 claim has a reasonable chance — roughly 45 to 60 percent — of surviving in some form against Anthropic and OpenAI, the defendants alleged to have made repeated confirmatory statements and to have participated in the working group. The odds are lower for Google (30 to 45 percent), given the caveated endorsement, and lowest for SpaceXAI (15 to 30 percent), which is alleged to have done nothing beyond a three-word post. The Clayton Act § 4 damages theory is more likely than not to be trimmed; the § 16 injunctive claim is more likely than not to survive if agreement is plausibly pleaded.
Leave to amend is near-certain after a first dismissal. Ninth Circuit practice is liberal, and futility findings typically follow a second failed attempt, as in Kendall, where amendment was denied only after plaintiffs had been “granted leave to amend once” and given discovery yet repeated “the same defects” (518 F.3d 1042).
The timeline is long. Rule 12 briefing should conclude within four to six months of service, with a ruling six to twelve months out. Copycat filings, whether related under Civil L.R. 3-12 or, if filed in other districts, centralized by the Judicial Panel on Multidistrict Litigation, could add a further three to six months. If claims survive, class certification would likely be decided in 2028 or 2029, putting any trial in 2029 at the earliest and more likely later.
Plausible end states, in rough order of likelihood: a narrowed case proceeding to discovery against fewer than four defendants (35 to 45 percent); dismissal with prejudice after amendment (25 to 35 percent); settlement, likely structured as conduct commitments given the damages problem (20 to 30 percent conditional on surviving Rule 12); and trial (under 10 percent). The single largest source of uncertainty lies outside the courtroom. A Department of Justice business review letter, an FTC statement, or legislation creating a safety-coordination exemption — which the essay itself is alleged to have requested (¶ 69) — would reshape the case more than any party’s briefing.
What to take from this
The safety debate will dominate the headlines, but the case will turn on three narrower questions: whether public endorsement of a proposal is an agreement, whether injury can be pleaded before effects appear, and whether the plaintiffs are bound to arbitrate. Each is an open question on these facts. That, more than any particular outcome, is why the case merits attention.
This analysis is based on the complaint and verified authority as of September 29, 2026. It was drafted using Midpage in Claude, with every case, statute and quotation retrieved and verified through Midpage. It is not legal advice, and no party’s conduct is characterized as unlawful.