OpenAI asked Congress to say whether a coordinated AI slowdown breaks antitrust law

The argument is about which act would be the illegal one. Felstead's answer depends on the paperwork. It turns "entirely on the precise details of any agreement," he wrote, and "even if most safety collaborations would ultimately survive antitrust scrutiny, legal uncertainty can act as a powerful deterrent" ^[6][7]. John Schulman drew the line somewhere else. "They'll cite antitrust, but that's fake," the OpenAI cofounder, now chief scientist at the rival lab Thinking Machines, wrote on X earlier this week, adding that "antitrust prohibits certain agreements, but not from jointly developing a proposal" ^[15].

The person this ask is really for is a general counsel, and counsel sorts this work into two piles. One is process: shared safety bars, evaluations, incident reporting. The other is output: who ships which model, and when. Caleb Knapp, director of government affairs at the AI Policy Network, which endorsed the July bill, says it would create legal channels for AI labs to work together to address safety and security incidents ^[10]. WIRED describes the bill's scope as coordination on security and safety work ^[8]. Nothing in that description says whether a joint decision about release timing counts as safety work.

Timing is the other half of the operator's problem. The House version has not been taken up in the Judiciary Committee ^[9], and Knapp says Congress has a "growing appetite to get something done" on AI safety while enactment may have to wait until after the midterms ^[11]. Any voluntary slowdown attempted before then runs without a statute behind it ^[1].

There is a competing read inside the industry, and WIRED reports it directly: another camp of AI leaders argues the antitrust worry is convenient cover for reasons that have nothing to do with legal liability ^[13]. The reasons WIRED gives are competition to capture a slice of the nascent frontier-model market, executives who share the Trump administration's view that staying ahead of China is a national security matter, and developers who hold vastly different opinions about how to build safe AI in the first place ^[14]. OpenAI did not respond to a request for comment ahead of publication ^[12].

So two axes, for whoever has to write the commitment. Unilateral or joint, and process or output. Unilateral process work needs nobody's sign-off and no new law. Joint process work is what the bill was drafted to protect. Joint output commitments, the pacing agreements, are the cell where Felstead's uncertainty actually bites, and they are also the only cell that requires asking Congress anything. Pachocki's own near-term forecast sits in the unilateral column: he expects "voluntary slowdowns to become commonplace until shared safety bars are established" ^[4]. WIRED's account names no lab that has slowed anything down yet.