Slow down, get sued: AI's safety pact hits antitrust court

Four people who pay for ChatGPT, Claude, Grok or Gemini walked into the U.S. District Court for the Northern District of California and accused the labs of fixing the pace of the future. The claim: Anthropic, OpenAI, SpaceXAI and Google coordinated an illegal slowdown, and paying subscribers got a thinner product for the same monthly fee. They want a nationwide class.

The alleged smoking gun is entirely public. On Sept. 12, Dario Amodei published an essay urging industry-wide cooperation on decelerating capability work in favor of safety — and Altman, Musk and Hassabis each endorsed it the same day. The suit traces the coordination back further, to a July joint statement about "intense competitive pressure not to unilaterally slow." Amodei had flagged this exact hazard himself, asking Washington to mediate or grant a narrow antitrust waiver for safety conversations; Altman answered that nobody needs to wait for one. Lead attorney Nick Rowley's pitch is the mirror image — AI safety shouldn't be "controlled by private agreements" either.

The timing is the punchline. Five days after the essay, Anthropic published its first R&D Automation Index: Claude now leads 26% of the company's own AI R&D, up from under 1% in February, with roughly 30,000 internal agents running at once. Nothing measured hit full autonomy — and every number in it is self-graded and externally unverified.

Why it matters: if "we agreed to go slower" is illegal, the only safety coordination left is the kind a regulator orders — and nobody has written that rule yet.