Four Subscribers Say AI Rivals Made a Deal to Slow Down

Four people who pay for ChatGPT, Claude, Grok and Gemini say Anthropic, OpenAI, SpaceXAI and Google cut a deal to slow AI on purpose.

Four people who pay for ChatGPT, Claude, Grok and Gemini say Anthropic, OpenAI, SpaceXAI and Google cut a deal to slow AI on purpose. They filed a federal antitrust lawsuit Friday in the Northern District of California. Their claim is blunt. The companies agreed to hold back. That agreement, they say, shrank the value of the subscriptions they pay for every month.

The suit does not argue that a single lab should not choose caution on its own. It argues that rivals cannot choose caution together. Antitrust law, the plaintiffs write, does not let competitors decide among themselves that competition is too dangerous. The shortcut they took, the complaint says, was to swap individual accountability for a collective pause.

The public moment at the center of the case came on September 12. Anthropic CEO Dario Amodei posted an essay calling for industrywide coordination to slow the pace of AI capability gains in favor of safety. OpenAI CEO Sam Altman, SpaceXAI CEO Elon Musk and Google DeepMind co-founder and chair Demis Hassabis each responded that same day in agreement. The plaintiffs point to those responses as evidence of a pact.

But the complaint also says the coordination started earlier. It points to a July 2026 statement signed by senior staff at several leading labs. That statement acknowledged intense competitive pressure not to unilaterally slow development. It urged the government to back a worldwide push to decelerate automated AI progress. The plaintiffs say that is the seed of the agreement they are challenging now.

What the lawsuit claims

The four named plaintiffs are paying subscribers to the four services. Their lawyers are seeking class-action status for other paid users of ChatGPT, Claude, Grok and Gemini across the United States. The complaint contends that the alleged coordination damaged consumers by diminishing the value of their paid AI subscriptions. It alleges a violation of Section 1 of the Sherman Act, the 1890 law that bars competitors from agreeing to restrain trade.

The suit draws a line that matters. It does not challenge companies individually choosing to slow their own AI development. It challenges the act of coordinating that restraint with rivals. The plaintiffs argue that antitrust law prohibits using the shortcut of agreeing to substitute collective restraint for individual accountability. In their reading, a competitive market can deliver both responsibility and progress. A private pact among direct competitors cannot.

The complaint also notes the companies have not denied the coordination. As of the latest reports, none of the four defendants had publicly commented on the lawsuit. The case is at the pleading stage. The court has not determined that the companies reached an agreement or violated antitrust laws. The plaintiffs still must clear class certification before the suit can move forward in earnest.

Why this lands hard

If the allegations hold up, the story is not just about AI safety. It is about who pays when rivals agree to move slower together. Subscribers pay monthly fees expecting their tools to get better. The plaintiffs say those fees now buy less because the labs synchronized their pace of innovation. That is the human problem at the core of the filing. You pay the same. You get less. And the reason, the complaint says, is a deal among competitors.

There is another layer. Amodei’s essay warned that swarms of rogue AI agents could take over the internet in as little as six months. That is a scary line. It is easy to see why leaders would want to slow down. But the plaintiffs say the law still draws a boundary. Companies can ask Congress or the White House for regulation. They can seek an antitrust exemption for safety coordination. What they cannot do, the complaint argues, is make that call on their own with their rivals.

I keep coming back to the phrase in the filing. Collective restraint instead of individual accountability. It sounds clean on paper. It feels different when you are the one paying the bill. The suit does not say AI should race without guardrails. It says guardrails cannot be built on a private handshake between competitors. That distinction is the whole case.

None of this is settled. The companies may argue the slowdown was independent or necessary for safety. They may argue the public statements were not a binding agreement. They may argue the plaintiffs lack standing or that the class cannot be certified. All of that will play out in court. For now, the record shows four subscribers, one complaint, and a claim that the value of their subscriptions dropped after a coordinated slowdown.

And then there is the quiet part. The part where you read the warning about rogue agents and you feel the pull to slow down. You read the antitrust line and you feel the pull to keep competition honest. Both pulls are real. Both matter. The lawsuit does not solve that tension. It just puts it on the docket and asks a judge to draw the line.