Antitrust suit tests whether AI labs can jointly agree to slow down

A proposed class action filed in the Northern District of California on 18 September puts a price tag on AI safety coordination: if a judge agrees that rival labs conspired to hold back product improvements, the industry's most prominent public safety pledge could become a textbook Sherman Act violation.

The case, Buist et al. v. Anthropic PBC et al., was brought by four paying subscribers to Claude, ChatGPT, Grok and Gemini, led by attorney Nick Rowley. It targets Anthropic, OpenAI, Google and SpaceXAI over events on 12 September, when Anthropic chief executive Dario Amodei published "We Must Pace the Frontier," calling for industry-wide slowing of capability gains. The complaint alleges that OpenAI's Sam Altman, SpaceXAI's Elon Musk and Google DeepMind's Demis Hassabis signalled agreement the same day, and cites an earlier July statement referencing "intense competitive pressure not to unilaterally slow" down as evidence the companies felt bound by rivals' moves rather than acting alone.

The legal theory is narrow but pointed: plaintiffs say they have no objection to any single company slowing its own development for safety reasons. Their claim is that four competitors agreeing together to restrict output is an unlawful restraint of trade. "The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous," the complaint states. They are seeking a liability finding and an injunction under Section 16 of the Clayton Act rather than damages.

The timing matters. Congress recently stripped a related antitrust shield from a defence bill, and the Justice Department has signalled that narrower cybersecurity cooperation between labs would not raise the same concerns — a distinction the plaintiffs will lean on to argue that broad "pace the frontier" coordination sits on the wrong side of the line. None of the four companies has responded to the suit, according to ABC News.

The suit forces a question AI labs have avoided answering directly: can they publicly coordinate on safety without triggering competition law designed for pricing and output cartels? If plaintiffs can show that Amodei's post functioned as a signal that rivals accepted, rather than a unilateral policy statement, the case could reshape how labs talk about pacing in public — pushing safety coordination into private channels or formal regulatory frameworks instead of open letters.

For the labs, exposure is reputational as much as legal at this stage: an injunction, not damages, is the relief sought, and none have yet had to answer the claims in court. But a live antitrust theory tying "responsible slowdown" rhetoric to unlawful output restriction changes the calculus for every future joint statement on capability limits, at Anthropic, OpenAI, Google DeepMind and elsewhere, according to Bloomberg Law.