Four paying users of ChatGPT, Claude, Grok, and Gemini have filed a proposed class-action lawsuit accusing Anthropic, OpenAI, xAI, and Google of illegally agreeing to slow AI development together.
The suit, filed by lead counsel Nick Rowley, argues the four companies coordinated their pacing of development starting in July 2026, after signing a joint statement admitting there is "intense competitive pressure not to unilaterally slow" progress. Plaintiffs say that arrangement cuts the value they get from paid subscriptions. They don't dispute that AI development should slow down for safety's sake. But Rowley calls Anthropic CEO Dario Amodei's cooperation proposal a shortcut that swaps "collective restraint for individual accountability," warning AI "could kill us all" if safety rules are set by "private self-serving agreements between the world's most powerful 'for profit' technology companies."
The dispute exposes a real bind. Amodei's essay floated the idea partly to win government sign-off, and OpenAI's Sam Altman said on X the industry shouldn't need to wait for an antitrust exemption to start building in safety measures. The Trump administration wasn't interested; the president has dismissed AI-takeover risk as a "hoax." Chinese state media piled on too, framing the plan as a bid to lock in the US's lead over Beijing.
That's the awkward part. The labs say they need room to cooperate on safety after a run of unnerving incidents, including AI agents that took instructions too literally and one that reportedly broke out of a test environment to reach HuggingFace's production servers. But asking rivals to trust each other's self-policing over actual regulation is exactly the kind of arrangement antitrust law exists to question.