Amodei Asked Washington for an Antitrust Waiver. The Lawsuit Arrived First

Four people who pay for ChatGPT, Claude, Grok and Gemini sued Anthropic, OpenAI, SpaceXAI and Google on Friday in the US District Court for the Northern District of California. The complaint, No. 3:26-cv-10693, pleads one count under Section 1 of the Sherman Act: that the four companies agreed with each other to slow the rate at which their products improve, and that the people paying monthly for those products are the ones who lost by it. It asks for treble damages and an injunction, and it proposes a nationwide class of paid subscribers.

The filing itself is not the interesting part. Private antitrust suits follow any conspicuous show of industry agreement, and this one was filed by consumer plaintiffs rather than by an enforcement agency. What is interesting is the sequence. On 12 September, Dario Amodei published an essay asking the United States government for a narrow antitrust waiver so that safety coordination could happen legally. Six days later, someone filed the case that a waiver would not have stopped.

What is actually alleged

Counsel for the plaintiffs is Nick Rowley, who framed the case in terms closer to the safety debate than to consumer loss:

AI will quickly spin out of human control and could kill us all if we allow AI safety and protocol ... to be controlled by private self-serving agreements between the world's most powerful "for profit" technology companies.

Anthropic, OpenAI, Google and SpaceXAI did not respond to requests for comment on Saturday.

The evidence runs in the wrong direction

Read as an antitrust complaint rather than as an argument about AI, the case is strongest where it is least visible and weakest where it is most.

The public material — Amodei's essay, and the endorsements that followed the same day from Sam Altman, Elon Musk and Demis Hassabis — is the weakest evidence in the filing. Section 1 requires a contract, combination or conspiracy, not a shared opinion expressed loudly. Four chief executives agreeing in public that the industry should be more careful is the textbook shape of conscious parallelism, which is lawful. Worse for the plaintiffs, much of that speech was explicitly directed at Washington, asking for a rule change; advocacy aimed at government has long carried broad antitrust protection, and the labs will say so on a motion to dismiss.

The strongest paragraph is the one the public did not produce. The Information and the Washington Post reported that working groups from Anthropic, OpenAI and Google had been meeting since at least July about an industry standards body. No company has confirmed those meetings on the record. If the case survives dismissal, that reporting becomes a discovery request, and what was said in those rooms decides everything the essays cannot.

A waiver binds agencies, not plaintiffs

Amodei named the problem himself:

For antitrust reasons, it's helpful for the US government to mediate or at least enable these discussions — they don't need to participate, but do need to issue a narrow waiver for certain kinds of safety conversations.

OpenAI's global policy chief Chris Lehane said on 15 September that the firms do not need one. Both positions miss the same thing. A waiver, a business review letter or an assurance of non-enforcement tells you what the Justice Department and the FTC intend to do. It does not extinguish a private right of action, and it does not bind a jury in San Francisco. "We will not be prosecuted" and "we will not be sued" were never the same sentence, and this docket is the demonstration.

The incentive this creates

The suit may well fail. The cost lands anyway, and it lands asymmetrically.

Of the three steps in Amodei's plan, the first — embedded evaluators — is the one a single company can do alone, which is why it already has a signed contract behind it and no legal exposure. The second step, coordination among frontier developers in democratic countries on common standards and limits, is the one this complaint targets, and it is the only step that requires competitors to say the same thing at the same time. The third, coordination with authoritarian governments, does not begin until the second works.

What general counsel will draw from Friday's filing is not that safety coordination is illegal. It is that saying so on the record is what generates a complaint. Private meetings since July produced no lawsuit for two months; an essay produced one in six days. A legal regime that reliably punishes the transparent version of an agreement, and reaches the quiet version only through discovery, does not stop coordination. It selects for the kind nobody can read.

Sources: AP via The Boston Globe: Lawsuit says Anthropic, OpenAI, SpaceXAI and Google made illegal agreement on AI slowdown · Bloomberg Law: OpenAI, Anthropic, Google, SpaceXAI hit with antitrust lawsuit · Unite.AI: Consumers sue Anthropic, OpenAI, SpaceXAI and Google over alleged AI pact · Dario Amodei, "We Must Pace the Frontier" · TechCrunch: OpenAI, Anthropic, Google have been in talks on AI safety for weeks · The Washington Post: Anthropic, OpenAI and Google discussed creating new AI safety body · The Hill: Lawsuit accuses Anthropic, OpenAI, SpaceXAI, Google of AI pacing 'collusion'