Does AI need an antitrust exemption so it doesn’t kill everyone????

Nilay Patel’s Decoder interview with Jonathan Kanter—ex–DOJ antitrust lead under Biden, now teaching at WashU and working tech policy at Carnegie Mellon—opens a two-part series on the future of business with a blunt question: do AI labs need antitrust exemptions so they can coordinate on safety instead of racing toward catastrophe?

The backdrop is familiar: safety resignations, researchers floating worse-than-10% odds that AI kills everyone, CEOs asking to slow development and seeking legal cover to collaborate. Critics call that regulatory capture, a cartel move, or a way to escape IPO-era investor pressure. Kanter’s record includes cases against Google, Apple, and Ticketmaster; he won Google and Ticketmaster, and the Apple case remains live under the Trump DOJ. Even David Sacks, a former Trump AI czar, approvingly amplified Lina Khan’s line that no antitrust exemption is needed.

We’re for useful AI that ships under real rules—not for a carve-out that lets the biggest labs set the pace together and call it survival.

Kanter’s car analogy is the clearest frame: inventing trucks without lanes, lights, stop signs, or speed limits. You need basic rules of the road. A generous read of “pace the frontier” is that labs fear speed plus no lines and believe doomsday is possible. A cynical read is that they are burning cash and want permission to slow without cratering valuations. Neither reading, on his account, earns an antitrust exemption.

Companies do not need a joint slowdown pact to deliver safer products—Boeing and Airbus do not get a cartel waiver to keep planes from falling out of the sky, and employee or agent liability still applies. Legitimate collaboration already exists: threat-sharing clearinghouses. Antitrust does not block that. Agreeing “we’re competing too hard, let’s all ease off” is the kind of coordination that can implicate the law.

Products liability recently bit Meta on Facebook and Instagram child safety and mental health—a settlement that took roughly a decade, which Kanter flags as the problem with waiting. Congress looks incapable; firms say they want regulation while lobbying against it. His sharpest line is also the simplest: if you believe the product will destroy humanity, don’t build it.

On geopolitics, he treats the China race argument as a convenient boogeyman—echoing how AT&T’s Cold War national-network pitch failed to stop the Reagan-era breakup. China already limits kids’ tech, censors, and runs state-influenced enterprises; “the only way to compete is unsafe products” is, in his words, ridiculous. He wants dynamic competition at home, domestic open-source and open-weights challengers, and clarity on consequences when products or agents cause real harm—mental health, competition, authenticity, IP turned into rules of the road—not calcified national champions that are hard to unwind.

I’m watching whether “safety coordination” stays in threat-sharing lanes or slides into collective slowdown talk that invites antitrust heat—and whether states keep moving while Congress stalls.

The Verge published the Decoder conversation September 19, 2026. Kanter also notes an earlier push for a ten-year state AI moratorium without federal rules, and argues states are currently better positioned than a gridlocked Congress to set enforceable baselines.