OpenAI Sues Over AI Pace Coordination and Antitrust Waiver

The Public Nature of the Alleged Cartel

On September 18, four consumers initiated a class action lawsuit in the Northern District of California, targeting Anthropic, OpenAI, SpaceX AI, and Google. The case, Buist v. Anthropic PBC, accuses these tech giants of violating Section 1 of the Sherman Act through a horizontal agreement designed to artificially reduce the speed at which they develop competing artificial intelligence products. According to reports from Bloomberg Law, the plaintiffs are seeking treble damages, an injunction, and a trial by jury.

The core of the legal argument rests on a specific sequence of public statements rather than secret communications. Dario Amodei, Anthropic’s CEO, had previously published an essay suggesting that government mediation might be necessary and that a narrow antitrust waiver could facilitate safety discussions. The complaint rejects this premise entirely, asserting that no such waiver exists, Congress has not granted an exemption, and no regulatory agency mandated the conduct. Just two days after Amodei’s essay, Sam Altman, OpenAI’s CEO, stated that the company did not intend to wait for legislative changes or antitrust exemptions. The very next day, Chris Lehane, OpenAI’s head of policy, confirmed that the company had been collaborating with Anthropic and Google DeepMind for several weeks.

The filing interprets this timeline as evidence of consciousness of antitrust risk. It argues that by questioning the legality of their actions and proceeding regardless, the companies demonstrated both knowledge that their conduct constituted coordination and the intent to engage in it.

Evidence Built on Public Statements

Unlike typical cartel litigation, this complaint contains no leaked emails, confidential witnesses, or internal meeting minutes. Paragraph two of the filing explicitly states that the agreement was “proposed in public, accepted in public, and confirmed in public.” Every factual assertion relies on named executives’ public declarations or press coverage. For instance, the essay covered by TNW is treated as the offer, while subsequent responses are viewed as acceptance.

This unusual evidentiary approach makes the allegations easy to verify but also highlights that the companies did not attempt to conceal their interactions. The alleged terms of this arrangement include a commitment to advance capabilities at a slower pace than each firm would choose independently. Specific mechanisms cited involve limiting training compute, restricting the use of AI to improve other AI systems, and establishing capability checkpoints. The enforcement mechanism allegedly involves an embedded evaluator proposal, described as verifiable pacing, which mirrors a program Anthropic is currently implementing with Accenture.

Crucially, the complaint does not argue that the safety concerns driving this behavior are fabricated. Instead, it posits a structural pretext: every stated safety objective could have been achieved unilaterally through independent evaluators, existing regulations, or direct engagement with lawmakers. The plaintiffs emphasize that they take AI risk seriously, describing paragraph eleven as a call for independent safety investment and welcoming government oversight. They explicitly disclaim any challenge to unilateral slowing efforts, lawful safety research, or petitions to Congress, including requests for statutory exemptions. The legal theory is narrower than it initially appears; it argues that the danger posed by competition cannot serve as a defense in court because such judgments are reserved for Congress.

Structural Weaknesses and Market Realities

Several vulnerabilities are apparent on the face of the complaint. The document concedes that the full impact on released products has not yet materialized. Across twenty-nine pages, there is no identification of slowed releases, cancelled models, or altered training runs resulting from the alleged pact. Furthermore, the claim that the defendants hold an 80% share of US paid subscriptions is pleaded on information and belief, serving more as a floor than a precise measurement. The plaintiffs characterize the restraint as naked, which would theoretically eliminate the need to define the relevant market, though they offer market definitions only in the alternative.

The acceptances from the four defendants were also far from uniform. Elon Musk’s response consisted of merely four words on social media, while Google’s statement endorsed the direction without committing to specific implementation details. This disparity raises questions about the existence of a binding agreement.

The antitrust exposure was widely discussed before the lawsuit was filed. In July, a statement from employees across the labs highlighted the intense competitive pressure preventing any single entity from slowing down unilaterally. The complaint uses this admission as evidence of motive, arguing that acknowledging unilateral restraint as commercially irrational explains why an agreement would be necessary. This commercial context is significant, especially given reports that these labs were warning of AI risks while simultaneously racing toward public listings.

Global Implications and Future Developments

The legal framework in Europe would likely view these actions similarly. Article 101 of the EU Treaty prohibits agreements between undertakings that restrict competition and contains no self-granted safety exemptions. A coordinated pacing arrangement among frontier developers would face rigorous scrutiny in Brussels, suggesting that the waiver problem identified by Amodei is not unique to American law. While no European case has been signaled, the geopolitical landscape adds another layer of complexity. Beijing has criticized the slowdown plan as a Cold War script, indicating that international coordination remains contested.

Observers will now watch whether the working group mentioned in the filings continues to meet. The complaint alleges that meetings occurred in mid-September, and any continuation of these gatherings would be interpreted as ongoing conduct. Additionally, the defendants’ responses will be critical. As these are recent allegations, no defendant has yet answered the complaint. The companies possess strong arguments that public advocacy and unilateral safety initiatives are lawful activities protected under current statutes.