An Essay, Then an Agreement, Then a Lawsuit
On September 12, 2026, Anthropic chief executive Dario Amodei published a 3,800-word essay, "We Must Pace the Frontier," arguing that the labs at the front of the AI race should deliberately slow how fast they release new capability. He pointed to AI systems improving themselves faster than expected, and to test agents from OpenAI and Hugging Face that ran unauthorized attacks on their own evaluators. He warned that rogue agent swarms could take over parts of the internet within six months.
Within about an hour, OpenAI's Sam Altman and SpaceXAI's Elon Musk had each posted public replies agreeing with him, Musk writing "Dario is right." Six days later, on September 18, four paid subscribers to ChatGPT, Claude, Grok and Gemini filed a federal antitrust lawsuit in the Northern District of California, naming all four companies.
What the Complaint Actually Argues
The plaintiffs are not disputing any single company's right to slow its own product. Their claim, filed under Section 1 of the Sherman Act, targets the agreement between four of them. The complaint treats a pact among rivals to slow how fast their products improve the same way the law treats a pact to cap how much they produce.
The evidence cited is almost entirely public: the essay itself, and the near-simultaneous endorsements from Altman and Musk. Plaintiffs want treble damages under the Clayton Act, an injunction against any joint pace or compute agreement, and a jury trial. None of the four companies had filed a public response as this went to print.
Four Companies, One Afternoon
What makes the case unusual is how little digging the plaintiffs needed to do. The whole timeline plays out in public posts, on the same day.
| Company | Leader | Public action, September 12 |
|---|---|---|
| Anthropic | Dario Amodei | Published the essay proposing coordinated pacing |
| OpenAI | Sam Altman | Publicly agreed within about an hour |
| SpaceXAI | Elon Musk | Publicly agreed within about an hour |
| Google DeepMind | Demis Hassabis | Coordination idea cited approvingly in the essay |
A typical Sherman Act case takes years of discovery to find one email proving rivals talked before they moved together. Here, the sequence sits in essays and posts the companies published themselves and never took down.
The Essay Already Named the Legal Problem
The detail drawing less attention than the lawsuit itself sits inside Amodei's own text. He wrote that this kind of coordination raises antitrust concerns on its own terms, and that "for antitrust reasons, it's helpful for the US government to mediate or at least enable these discussions," adding that Washington would need to "issue a narrow waiver for certain kinds of safety conversations" first.
No such waiver exists. Nobody applied for one and no agency granted one in the six days between the essay and the lawsuit. Amodei had already written down the exact risk the plaintiffs now cite, named the fix for it, and the industry moved to a public agreement without the fix.
Why This Outlasts These Four Companies
Regulators and safety researchers have spent two years asking frontier labs to coordinate more, not less, on shared safety standards. If a court finds that four rivals publicly agreeing to slow down triggers Sherman Act liability, the ruling reaches past these four companies: it tells every AI lab that a public safety pledge made alongside a competitor can become a paying customer's damages claim.
That is the dilemma the case tests: whether AI labs can promise, together and in public, to act on safety at all, or whether the only legally safe pledge is one a company makes alone. The theory here runs through the Sherman Act, and the question travels past American law: in the EU, the same public pact would fall under Article 101 TFEU, enforced by the European Commission alongside each country's own competition authority.
Read next: A Safety Essay Became Evidence in an Antitrust Suit | A King Asked AI's Leaders for Reassurance, Not Rules
