September 20, 2026
Anthropic, OpenAI, Google and xAI sued for agreeing to slow AI: the suit was filed by paid subscribers
On September 18, four AI leaders were accused of the opposite kind of collusion: not driving progress forward, but slowing it down.
Charles Buist pays for an AI subscription and on September 18 filed suit against four labs at once.
The lawsuit is a class action, with a nationwide class proposed. Buist and three other subscribers are suing Anthropic, OpenAI, Google, and xAI in the federal court for the Northern District of California under Section 1 of the Sherman Act, alleging a conspiracy in restraint of trade.
The trigger is stated plainly. On September 12, Dario Amodei published the essay “We Must Pace the Frontier,” arguing that the pace of model improvement should be slowed. Altman, Musk, and Hassabis publicly endorsed it. From this, the plaintiffs assembled a timeline: a working group of lab representatives in July, an OpenAI essay on options for an industry-wide slowdown on September 6, and confirmations of coordination on September 14–15.
Amodei explained why to slow down with numbers. He considers an internet takeover by a swarm of agents possible within a 6–12 month window, and would devote the 1–2 years gained by slowing down to work on interpretability.
The plaintiffs seek treble damages and a court order barring agreements on the pace of development, compute limits, and release dates. The lawsuit does not touch companies' independent safety decisions.
The market in the suit is defined narrowly: paid subscriptions to the smartest assistants, such as ChatGPT Plus at $20 a month. The four defendants control at least 80% of it.
Previously, labs sued over how quickly and on what data they trained models. Now they face the opposite allegation, and the injured party is the paying subscriber.
So far, nothing has slowed down. As of September 17, none of the four companies had limited either compute or release timelines. Anthropic gave external evaluators employee-level access and allowed them to publish conclusions without its edits; OpenAI promised the same without publishing terms, while xAI relies entirely on a verbal agreement.
Amodei foresaw the antitrust risk himself: in the essay, he asked the U.S. government to act as a mediator and grant a narrow exemption for lab discussions about safety.
The outcome of the lawsuit will determine whether labs can legally discuss a shared pace of development at all.
Source
