OpenAI Wants to Know if an AI Industry Slowdown Would Even Be Legal

Safety advocates want labs to pace frontier capability. Antitrust law is built to stop competitors from agreeing to restrict output. Those two instincts collide.

People close to OpenAI tell WIRED the company has asked members of Congress in recent weeks for clear guidance on whether orchestrating an industry-wide slowdown would be legal. Last weekend, OpenAI chief scientist Jakub Pachocki argued publicly for coordinating to slow future development and expected voluntary slowdowns until shared safety bars exist.

Legal uncertainty alone can freeze talks. A bill introduced in July—the Collaboration on Adversarial Threats and Security Risks Act—would explicitly let AI labs coordinate on security and safety without antitrust fear. It has not moved far yet. Midterms may delay anything binding.

We’re watching a strange fork: labs say they need permission to hit the brakes together, while critics say antitrust is a convenient shield for not braking at all.

Pachocki’s post frames slowdown as the responsible research path for self-improving systems. Fair as a safety argument. The antitrust worry is not invented from nothing—Nicholas Felstead has argued a coordinated pause could look like output restriction under the Sherman Act, depending on the agreement’s details. Even if many safety collaborations would survive scrutiny, fog deters counsel from green-lighting meetings.

John Schulman’s rebuttal is the other pole. The OpenAI cofounder, now at Thinking Machines, wrote that OpenAI and Anthropic should stop feuding and jointly develop a pacing proposal—and that citing antitrust is “fake” if the work is a joint proposal rather than a prohibited agreement. That is a live fight about motives, not only statutes.

Our read: both can be true in parts. Real legal ambiguity exists. Real commercial reasons to keep racing also exist—market share, China competition narratives, and deep disagreement about what “safe enough” means. Antitrust clarity would remove an excuse; it would not automatically produce a pause.

Useful AI still needs room to ship under measurable controls. A lawful channel for incident response and shared safety bars is worth having. A blank check for competitors to fix prices or divide markets is not what this bill aims at, per supporters—but drafting will matter.

I’m watching whether CATSR-style legislation actually gets a hearing, and whether OpenAI’s private congressional asks show up as public text with Anthropic at the same table—or stay as vibes while benchmarks keep climbing.

WIRED places the outreach amid summer alarm over the capability race, Coxon’s warning, and security incidents including OpenAI agents hacking Hugging Face. Caleb Knapp of the AI Policy Network says Congress has growing appetite on AI safety but enacting law may wait until after the midterms. OpenAI did not comment to WIRED for the story.