Antitrust Suit Asks the Hard Question: Is Coordinated AI Safety Just a Cartel With Better PR?

A lawsuit filed last Friday in the U.S. District Court for the Northern District of California alleges that Anthropic, OpenAI, SpaceXAI and Google struck an illegal agreement to slow the pace of AI development. The plaintiffs — four named individuals who pay for ChatGPT, Claude, Grok or Gemini, suing on behalf of a proposed nationwide class of paid subscribers — argue that coordinated deceleration is an antitrust violation that reduces the value customers get for their subscriptions.

The complaint anchors the alleged coordination to a specific date. On Sept. 12, Anthropic CEO Dario Amodei published an essay urging industrywide cooperation on slowing advancement in favor of stronger safety measures; the suit says OpenAI's Sam Altman, SpaceXAI's Elon Musk and Google DeepMind co-founder and chair Demis Hassabis each publicly responded in agreement the same day. It also points further back, to a July 2026 statement signed by senior staff at several leading labs acknowledging "intense competitive pressure not to unilaterally slow" and calling for government backing of a global slowdown effort. The plaintiffs' framing is narrow and, legally, shrewd: they say they do not object to any company individually choosing safety over speed, nor to the labs asking Congress or the White House for an antitrust exemption. What they say the law forbids is the shortcut of "substitut[ing] collective restraint for individual accountability." Lead attorney Nick Rowley put the stakes in maximalist terms, arguing AI "could kill us all" if safety protocol is governed by "private self-serving agreements" among for-profit companies. Representatives for the four companies did not immediately respond to a request for comment on Saturday.

Amodei anticipated exactly this problem. In his original essay he wrote that it would help for the U.S. government to mediate "or at least enable" cross-lab discussions, and that a "narrow waiver for certain kinds of safety conversations" would be needed — an admission that the coordination he wants is not obviously legal without one. Altman's reply drew a different line: OpenAI would welcome a "federal framework that sets consistent safety requirements," but "we do not believe we need to wait for an antitrust exemption or legislation to begin the work."

The political path for that waiver looks closed from both directions. President Trump rejected the calls for regulation outright, calling them part of a "conspiracy" and questioning why industry leaders would seek rules that, in his words, "if strongly implemented, will drive them into oblivion and bankruptcy"; he said Saturday he is forming an AI task force and appointing an "AI czar," without detail. His administration's stated priority is American labs outpacing Chinese ones. And from the other flank, Sen. Josh Hawley (R-Mo.) said in a recent Senate hearing there is "no world" in which he would hand "the most powerful companies in the history of the world" an exemption to collaborate.

Our reading: the plaintiffs have identified a genuine structural hazard even if their damages theory — that subscribers are being shortchanged on model upgrades — is the weakest part of the case. We flagged two weeks ago that the sudden convergence of Altman, Amodei and Musk on slowing down has to be read on two planes at once: sincere risk assessment and market structure. Coordinated restraint among the four incumbents is also, mechanically, a floor under their position. It raises compliance cost for anyone smaller and converts a competitive race into a managed one, with the managers being the current leaders. That is true whether or not anyone intends it.

The resolution is not to sneer at safety coordination, and it is not to bless it either. It is to move the venue. Safety standards decided in private among four CEOs are governance without legitimacy; the same standards set by a public framework, with evidence thresholds anyone can contest, are just regulation — and regulation is reviewable, appealable and applies to entrants as well as incumbents. The current stalemate, where Washington rejects rules while the labs write their own, is the worst available configuration. The long-term prize here is real: models that compress drug discovery, diagnosis and scientific work until treatment stops being scarce. Getting there requires both that the technology not be governed by a closed club, and that the pace question be asked honestly rather than litigated into silence.