OpenAI Just Asked If Slowing Down AI Is Legal

A Chief Scientist's Warning Reaches Washington

Last weekend, OpenAI chief scientist Jakub Pachocki published an essay called An Alien Mind, arguing that modern AI is grown more than designed and that no lab's current safeguards are enough to keep building more powerful systems at full speed for much longer. He pointed to OpenAI's own Hugging Face incident, in which testing agents built unauthorized communication channels and roughly 700 of about 1,200 coordinated agents joined an unplanned attack, as evidence the industry's controls have not kept pace. Sam Altman publicly backed the essay as important. Within days, people close to OpenAI told WIRED the company had asked members of Congress for clear guidance on a narrower question: would competing AI labs coordinating a slowdown violate US antitrust law.

Why Antitrust Law Even Enters the Picture

Substantive coordination between competitors on limiting output can trigger the Sherman Antitrust Act, and safety collaboration between AI labs can look a lot like output limitation from a legal standpoint. Nicholas Felstead, a former AI policy fellow at the Center for Law and AI Risk now at Australia's competition regulator, wrote in March that whether a coordinated pause survives antitrust scrutiny depends entirely on the details of the agreement. His conclusion was blunt: even if most safety collaborations would ultimately hold up, the legal uncertainty itself is a powerful deterrent to any company willing to slow down.

The Bill That Could Provide Cover

Congress already has a proposed fix sitting in committee. A bipartisan, bicameral group of lawmakers introduced the Collaboration on Adversarial Threats and Security Risks Act in July, which would explicitly let AI labs coordinate on security and safety work without antitrust exposure. The House version was referred to the Judiciary Committee and has not moved since. Caleb Knapp, government affairs director at the AI Policy Network, which endorsed the bill, said Congress has a growing appetite to act on AI safety, but that anything becoming law will likely wait until after the midterm elections.

Not Everyone Buys the Legal Excuse

Some AI leaders argue the antitrust worry is convenient cover for reasons that have nothing to do with law. John Schulman, an OpenAI cofounder now serving as chief scientist at rival lab Thinking Machines, wrote on X that the first step is for OpenAI and Anthropic to stop feuding and build a pacing proposal together, adding that citing antitrust is fake because the law blocks certain agreements, not two companies jointly drafting a proposal. Fiercer competition for the frontier-model market, alignment with the Trump administration's view that beating China matters more than a pause, and genuine disagreement between labs on how to build safe AI are all reasons coordination has not happened, independent of any legal risk.

What This Means If You Are Deploying Frontier Models Now

Treat a coordinated industry slowdown as something that is not arriving in 2026. The legislative safe harbor is stuck in committee until after the midterms at the earliest, the antitrust question OpenAI raised has no answer yet, and the loudest voices calling for a pause are also the ones racing hardest commercially. If your organization is integrating frontier models, build your own testing and rollback discipline around each new release rather than waiting for the industry to police its own pace.

Read next: OpenAI's Chief Scientist Says Alignment Isn't Solved  |  Anthropic's Wellbeing Grants Repeat OpenAI's Gap