A class action lawsuit filed September 18 in the U.S. District Court for the Northern District of California accuses Anthropic, OpenAI, SpaceXAI And Google of having violated antitrust regulations by coordinating to slow down the development of their respective models. The appeal is filed by four individuals, between Florida and California, who pay for a subscription to one of the four chatbots: the lawsuit aims to extend to an entire class of subscribers nationwide.
The file is based on the Sherman ActAmerica’s antitrust law of 1890, and seeks treble damages under the Clayton Act as well as an injunction against future horizontal development pace agreements. According to the lawyers, the synchronization between the four companies took concrete shape on September 12, when the founder of Anthropic Dario Amodei published an essay calling on the industry to collectively slow down in the name of safety. On the same day, Sam Altman (OpenAI), Elon Musk (SpaceXAI) e Demis Hassabis (Google DeepMind) responded publicly as a sign of support and there was immediate talk of a cartel.
According to the prosecution, however, the agreement did not arise that day. The lawsuit recalls a statement signed in July by top executives of the main laboratories, which openly acknowledged “intense competitive pressure not to unilaterally slow down” development. An admission which, according to the lawyers, demonstrates the premeditation of the agreement then made public in September.
The accusation: “A sign disguised as a security issue”
The lawyer Nicholas Rowleyhead of the legal college together with colleagues from the Trial Lawyers for Justice firm, does not deny that artificial intelligence should be slowed down for safety reasons. The disputed point is who decides how to do it: Amodei’s proposal is defined as one “shortcut” which “replaces individual responsibility with collective restraint agreed upon among the most powerful tech companies on the planet.” Rowley is clear: leaving such crucial safety protocols dependent on private agreements between for-profit entities risks allowing AI to escape human control, not preventing it.
The White House already has rejected Amodei’s approach: President Trump has publicly dismissed AI-related existential risk scenarios as a “hoax.” Chinese state media also intervened, defining the initiative as an agreement whose “admission criteria were written even before the guest list was drawn up”, in an attempt to slow down China rather than the entire sector: in his essay Amodei explicitly cited the desire to widen the US advantage over Beijing.
Doubts remain regarding the legal validity of the case. Critics of the approach point out that no subscription contractually guarantees a constant pace of improvement of the models, and that companies could invoke a public interest exception linked to security. Anthropic, OpenAI, Google And SpaceXAI have not yet responded publicly to the allegations contained in the complaint.
