Antitrust lawsuit over calls to slow AI development

On 18 September 2026, a lawsuit, Buist v. Anthropic PBC, was filed in the U.S. District Court for the Northern District of California: four paying subscribers accuse the largest AI companies of collusion. The grounds are unusual for an antitrust case. The plaintiffs are unhappy that competitors are publicly calling for technology development to be slowed, and consider such coordinated behavior a violation of the rules of the game.

The lawsuit came during a week when everyone was talking about slowing down: from a Fields Medalist to the White House chief, in the opposite direction.

What exactly is in the lawsuit

Case No. 3:26-cv-10693, filed on 18 September in San Francisco. There are four defendants: Anthropic, OpenAI, SpaceXAI, through which Elon Musk runs Grok, and Google. The plaintiffs are Charles Buist and Nick Spetsas from Florida, and Shayne Hunt and Christina Bullock from California. All four pay for subscriptions to the listed services. They are represented by Nicholas Rowley of Trial Lawyers for Justice.

The legal theory is based on the Clayton Act and Sherman Act § 1. The plaintiffs seek treble damages, an injunction, and a jury trial, and also intend to turn the case into a class action: subscribers to ChatGPT, Claude, Grok, and Gemini nationwide could join it.

The economic harm is stated simply. If the four leading competitors simultaneously slow the pace of development, users will receive fewer updates and improvements for the same money, meaning the value of paid subscriptions will fall. At the same time, the plaintiffs' attorney publicly agrees that AI needs serious safety measures: in his view, the rules should be established openly by the government and within the law, rather than by companies among themselves.

The grounds: a week of calls to stop

Dario Amodei spoke first. In early September, he published an essay calling on the largest companies to coordinate across the industry and proposing that the pace of development of advanced systems be reduced: in his argument, rapid progress without additional safety measures creates serious risks. Elon Musk, Sam Altman, and Demis Hassabis publicly supported him—that is, the heads of three competing laboratories. In the plaintiffs' view, this agreement is precisely what points to an illegal arrangement.

A little later, mathematician Terence Tao, a Fields Medalist, spoke out: he called the pace of development insane and expressed the situation with the phrase “We have to slow down.” Tao's argument is based not on ethics but on mathematics: AI development resembles a nonlinear system in which a small change produces an unpredictable result, and at such a speed of dissemination, neither researchers nor society has time to examine the consequences. He separately noted that the problem lies not only in the technologies themselves but also in the speed at which they spread.

Counterposition: the administration opposes restrictions

On 19 September, according to The Washington Post, Donald Trump announced plans to create an AI Force and appoint a separate official responsible for AI development, whom the media have already dubbed the “AI czar.” The president explicitly opposed attempts to slow the technology: the administration will not hinder the industry's growth, and, according to him, the illegal use of AI should be addressed through the existing criminal and civil legal system. He drew the comparison from his own first term: the future AI forces are being compared with the U.S. Space Force. Leadership in AI also remains part of the technological rivalry with China. The White House has not yet specified whether the structure will be military or civilian.

What is already regulated: the example of schools

While the laws are being debated, the first strict rules are arriving through contracts. On 9 September, it became known that Microsoft had reached an agreement with the AFT teachers' union, which the UFT joined. According to the AP, the terms are legally binding and take effect on 1 November 2026 for all schools with Microsoft contracts:

• chat companions designed to create an emotional attachment to or dependence on the system in a child are prohibited;
• student and teacher data may not be used to train models, except for safety purposes, advertising, or the development of new products;
• compliance is verified by independent auditors, not by the company itself.

This may be a more telling precedent than the lawsuit. A court case drags on for years, while the restriction on capturing children's attention takes effect in November and applies to all school contracts of the leading vendor.

What follows from this

• Follow Buist v. Anthropic: if the court allows the class action, the number of defendants and subscribers in the group may grow, and the precedent would then affect anyone who pays for subscriptions.
• Do not rely on public industry calls as regulation. Until industry statements have the force of law, only laws and contracts work.
• Check what your contract with the vendor says about data. The Microsoft school example shows the formula: a ban on specific functions, a ban on data use, and independent auditing.
• For businesses, the risk is mutual. Both slowing down and an anticompetitive agreement harm subscribers, so it makes sense to measure a provider's resilience not by management rhetoric but by the pace of patches and releases.