OpenAI, Anthropic and Google Push Frontier AI Slowdown as Critics Cry Cartel

OpenAI, Anthropic and Google have jointly floated a slowdown in frontier AI development, framing the proposal as a safety measure. The pitch lands at a moment when the three companies already hold most of the compute, talent and capital needed to train the largest models, and it has drawn immediate resistance from competitors and political figures who say the timing is not a coincidence.

The sharpest criticism came from Cohere chief executive Aidan Gomez, who described the effort as a cartel under a different label, built to keep smaller players out of the top tier of model development. His argument is straightforward: when the companies best positioned to comply with a slowdown are the ones writing the rules, the compliance burden falls hardest on everyone else. Cohere, which sells enterprise language models, competes directly with the labs backing the proposal.

The political reaction has been just as pointed. The White House and President Trump have come out against the slowdown, putting the current administration at odds with the labs on an issue that touches national competitiveness. That opposition matters because any serious brake on frontier training would likely need government cooperation, whether through procurement rules, export controls or enforcement of safety standards. Without it, the proposal reads more like a voluntary pledge than a policy.

The dispute exposes a split that has been forming for over a year. Safety advocates inside the big labs argue that capabilities are advancing faster than the ability to test and control them, and that a coordinated pause is the only way to buy time. Smaller labs, open-source groups and investors counter that slowing the leaders simply freezes the market in place, since any company that ignores the call gains ground while the signatories hold back.

What happens next depends on whether the slowdown gains any formal mechanism. A statement without verification, auditing or penalties is easy to sign and easy to abandon, and critics will point to that gap. Watch whether Congress or the administration proposes its own rules for frontier training, and whether any lab outside the big three signs on. The answer will show whether this is a genuine safety effort or a competitive maneuver wearing safety's clothes.