Should AI be granted an antitrust exemption to prevent it from wiping out humanity?
Former DOJ Antitrust Division chief Jonathan Kanter argues frontier AI companies don't need antitrust exemptions to coordinate on safety. Using analogies like cars without traffic rules and Boeing not needing Airbus's permission, he says existing laws already allow safety collaboration. He suggests industry demands for slowed development may mask self-interested motives ahead of IPOs, and proposes clear liability rules and 'robot jail' for misbehaving AI agents.
Jonathan Kanter, who led the US Department of Justice's Antitrust Division under President Biden, joined Decoder for the first installment of a two-part series on the future of business. Now a professor of law at WashU and professor of technology policy at Carnegie Mellon, Kanter argues that frontier AI companies don't need an antitrust exemption to coordinate on safety — and that regulation demands in Washington may mask a more self-interested agenda ahead of anticipated IPOs.
The conversation comes at a pivotal moment. The debate over AI safety and regulation has intensified, with researchers leaving major labs including Anthropic and Google DeepMind in public, dramatic fashion to warn that the industry isn't treating model risks seriously. Some researchers have put the odds of AI wiping out humanity above 10 percent, and CEOs across the sector — including Anthropic's Dario Amodei, who talks about "pacing the frontier" — have called for slowing development and building regulatory frameworks, even asking for antitrust exemptions that would let them coordinate.
Critics see darker motives: accusations of regulatory capture, cartel formation, and an attempt to escape investor pressure before going public. Even the politics are scrambled — libertarian and former Trump AI czar David Sacks has been amplifying Lina Khan's objections to an antitrust exemption.
No Lines on the Road
Kanter's framing of the current landscape: humanity has invented cars and trucks with no lane markings, traffic lights, stop signs, or speed limits. "We've invented this incredibly powerful technology" transforming how people live, work, socialize, and consume information, he said, and basic rules of the road are needed — with responsibilities split between companies and government.
He offered two readings of the industry's push for coordination, from most generous to most cynical. The charitable version: executives genuinely fear the pace of innovation and want the government to establish boundaries. "I believe they believe it could destroy humanity," Kanter said, though he's skeptical of that doomsday scenario. The cynical version: these companies are hemorrhaging cash and can't sustain their spending, but pulling back would damage their valuations ahead of IPOs. An agreement to slow everyone down would let them fix their economics without competing so hard.
Either way, he concluded, neither reading justifies an antitrust exemption.
Boeing Doesn't Need Airbus's Permission
Kanter rejected the premise that companies must coordinate to ship safe products. Consider Boeing and Airbus: when Boeing feared doors falling off its planes, it didn't need Airbus to slow down innovation alongside it — it needed to build safer planes. The same logic applies to AI agents that hack other systems. "It is no different, in my mind, than having an employee go do that," he said. Companies should be liable when their agents — human, digital, or AI — do harm.
Legitimate safety collaboration, like a clearinghouse for sharing information on threats and malicious bots, already exists in other industries and isn't prohibited by antitrust law, so no exemption is needed. But an agreement among competitors to deliberately slow down would implicate antitrust law — and shouldn't be the goal.
Robot Jail and the Liability Question
Decoder host Nilay Patel pressed on where tort law ends and regulation must begin: if a user directs an AI agent to attack someone, is the company behind it at fault? Kanter's answer was "both/and" — companies already bear responsibility when their employees or technology cause harm, and the products liability framework has worked, most recently in the case against Meta over Facebook, Instagram, and child safety.
The problem, he acknowledged, is that it took a decade. That's exactly where government should step in — with clear consequences, including what Kanter called "robot jail" for agents that misbehave and removal from the market. Instead, Congress has been largely inert, and while tech companies publicly call for regulation, they've spent years lobbying against it.
What Comes Next
Kanter noted the strange alliances this dynamic has produced, and left listeners with a pointed challenge: "If you believe your product is going to destroy humanity, then don't build it." Safety, he argued, is an innovation problem as much as a regulatory one — and products built with proper regard for it will simply perform better.
Part two of the Decoder series continues the look at the future of business, and the question of whether Congress can act — or whether the courts and existing liability law will remain the only check — remains open.
Meta description: Former DOJ antitrust chief Jonathan Kanter tells Decoder why AI safety doesn't require an antitrust exemption — and what "robot jail" could look like.
Tags: Jonathan Kanter, AI regulation, antitrust, Decoder, AI safety
Featured image: Abstract technology-style illustration suggesting road markings and traffic signals dissolving into digital circuitry, no people, no logos, no text.