Four AI Labs Agreed to Slow Down. Seven Days Later They Were Sued.
The essay went up on 12 September. The complaint was filed on the 19th.
In between, three rival chief executives publicly agreed with it.
Dario Amodei had argued that AI labs should coordinate to slow the pace of development in favour of safety. Sam Altman, Elon Musk and Demis Hassabis said much the same thing the same day.
Seven days later, four of their paying customers took them to federal court.
What the Lawsuit Says
It was filed on Friday 19 September in the US District Court for the Northern District of California.
The four named plaintiffs all pay for ChatGPT, Claude, Grok or Gemini. They are asking to represent a nationwide class of everyone who pays for these products. Their lead attorney is Nick Rowley.
The defendants are Anthropic, OpenAI, SpaceXAI and Google.
The claim itself is short. Competitors agreed among themselves to compete less hard. Paying customers got less product for the same money.
“The antitrust laws do not permit competitors to decide among themselves that competition is too dangerous.”
That one sentence from the filing is the entire case.
Why This Is Genuinely Awkward
Safety coordination and an illegal cartel look uncomfortably similar from the outside.
- Both involve competitors talking to each other. Antitrust law cares a great deal about that.
- Both reduce output. A slower model is less product than a faster one.
- Both were agreed in public here. Which makes them very easy to document.
The openness that made it credible as a safety move is the same thing that makes it simple to sue over.
Get Marlon Steiner’s stories in your inbox
Join Medium for free to get updates from this writer.
An essay and four public endorsements are not a smoking gun found in someone’s inbox. They are the front page.
What the Companies Have Said
Nothing, so far. None of the four responded to press requests when the story broke.
Worth holding on to: this is one filing by four individuals. It is not a ruling, a finding, or a regulator taking action.
A lawsuit is an allegation with a filing fee attached.
Plenty of class actions in this space have gone nowhere. This one has an unusually clean factual record behind it, which is the part that makes it interesting rather than routine.
Why It Matters If You Build on These Models
- If pacing holds, capability improves more slowly and your roadmap stretches with it.
- If the suit lands, coordinating on safety becomes legally risky and every lab decides alone.
- Either way, the assumption that model capability keeps compounding at the current rate is now a live question rather than a background fact.
The Wrap-Up
Four companies tried to agree that the race was getting dangerous, and a court will now decide whether they were allowed to.
Here is your action step. If anything you are building assumes next year’s models are much better than this year’s, write down what happens if they are not. That scenario stopped being hypothetical this week.
thank you for reading :)
This article runs on coffee. Refuel me here ☕️
