AI slowdown lawsuit targets ChatGPT, Claude, Grok and Gemini subscriptions

Four paying AI subscribers have sued Anthropic, OpenAI, SpaceXAI and Google, alleging that the companies agreed to slow improvements to their competing AI products while continuing to charge for premium access. The [29‑page complaint filed September 18](https://chatgptiseatingtheworld.com/wp‑content/uploads/2026/09/Buist_et_al_v_Anthropic_PBC_-Sept‑18‑2026.pdf) frames public calls for shared AI safety measures as an antitrust agreement. It is a newly filed allegation, not a finding that the companies coordinated their product roadmaps or that subscribers have already suffered a measured loss. The case has a direct question for people who pay for ChatGPT, Claude, Grok or Gemini: did a common approach to AI safety replace the pressure each rival would otherwise face to ship a more capable product? The plaintiffs say yes. Their filing also identifies the evidence they would need to test that claim, including internal development plans, compute allocations, release schedules and subscription data. None of those records is established by the public complaint itself.

The plaintiffs trace their theory to Anthropic CEO Dario Amodei’s September 12 essay, [“We Must Pace the Frontier”](https://darioamodei.com/post/we‑must‑pace‑the‑frontier). Amodei proposed three steps. First, outside evaluators would receive sustained access to a company’s models and training work; Anthropic said it would adopt that step on its own. Second, companies in democratic countries would work toward common safety standards and limits on unchecked progress. Third, governments would seek wider international coordination. The essay says some forms of company coordination would be legally difficult and require government support. The [complaint’s account of the following statements](https://chatgptiseatingtheworld.com/wp‑content/uploads/2026/09/Buist_et_al_v_Anthropic_PBC_-Sept‑18‑2026.pdf) is more aggressive than simply saying executives liked the essay. It treats Elon Musk’s approval, Sam Altman’s support for pacing and Demis Hassabis’s endorsement of the general direction as acceptances of a common restraint. It also points to later comments and reported talks among OpenAI, Anthropic and Google DeepMind about safety standards. Those statements and talks are central to the plaintiffs’ case; whether they establish an unlawful agreement to restrict competition is a question the court has not decided. That distinction matters because Amodei’s first step and his proposed second step have different practical meanings. A company can decide independently to let evaluators inspect its own systems. A common limit on how fast competing companies train or release models would require coordination. The filing argues that the public responses crossed that line, while the original essay describes the industry‑wide step as a proposal requiring further legal and government work. [Associated Press reporting before the lawsuit](https://apnews.com/article/ai‑slowdown‑anthropic‑openai‑meta‑nvidia‑1d9615931af28a83cb97489178e90f2d) also documented disagreement elsewhere in the industry: Meta and Nvidia executives pushed back on collective pacing. Public alignment among some leaders is therefore a narrower fact than an industry‑wide slowdown in operation.

The complaint focuses on a quality‑adjusted overcharge rather than documenting a specific nominal price rise. It argues that a subscription’s value depends partly on improvements in reasoning, coding, agents, speed and other capabilities, and that buyers would get less improvement for the same payment if rivals agreed to restrain their development. This is the part of the lawsuit that connects executive safety discussions to individual AI bills.

The filing contains an important limit on its own evidence. In paragraph 110, it says the full effect on released products has not yet appeared, because the alleged agreement is recent and model‑development cycles take months. It asserts, on information and belief, that incentives and some internal decisions have changed. Paragraph 118 then says any damages would need to be proved through discovery and expert analysis of plans, spending, capabilities, prices and customer behavior. The complaint does not identify a particular postponed model release or calculate an amount each subscriber supposedly overpaid.

That leaves two separate tests ahead. The plaintiffs would have to show more than executives discussing common safety standards: they would need to establish the alleged agreement and its scope. They would also need to connect it to a reduction in what subscribers received, compared with a credible account of what the four firms would have released without it. Recent research showing that Claude now performs a larger share of Anthropic’s own AI development work explains why the pace of capability development has become a live industry issue; it does not establish the counterfactual in this lawsuit.

The complaint asks to represent people in the United States who bought a paid individual consumer subscription directly from one or more defendants beginning September 12, 2026. It lists premium access to ChatGPT, Claude, Grok and Gemini and proposes a separate subscriber subclass for each defendant. That wording does not automatically include free users, every business or API account, or people who purchased through an intermediary. Nor has a judge certified the proposed class. The plaintiffs say they may revise its boundaries if discovery reveals that any agreement began earlier.

The requested remedies include damages and an order stopping the alleged coordination. They are requests, not an award or an injunction already in force. A filed complaint also does not certify a nationwide class. The next consequential questions are whether the defendants dispute the claimed agreement, what records the parties can obtain, and whether the plaintiffs can show that any coordinated action changed products or subscription value.

The Associated Press reported September 19 that representatives of the four companies had not immediately responded to its request for comment. Until responses and evidence enter the record, the useful distinction for subscribers is simple but consequential: the suit challenges an alleged shared restraint on product improvement, while the publicly available filing has yet to demonstrate that an actual release was delayed or a subscription lost measurable value.