Kanter: AI labs don't need an antitrust exemption for safety

The Verge's Decoder podcast ran the first of a two-part interview with Jonathan Kanter, the former antitrust chief of the US Department of Justice under the Biden administration, now a professor of law at WashU and of technology policy at Carnegie Mellon. The conversation covers whether AI companies should get an antitrust exemption so they can coordinate on safety, a request the host says CEOs across the industry have floated alongside broader calls to slow development. The host frames the backdrop: researchers at major labs including Anthropic and Google DeepMind have quit publicly, arguing models pose real threats and that safety isn't being taken seriously; unnamed researchers have put the chance AI kills everyone at greater than 10 percent. Dario Amodei of Anthropic has used the phrase 'pace the frontier' to describe slowing down. The host lays out three possible motives behind the exemption request: genuine fear of catastrophic risk, a cynical bid to relieve competitive and cash-burn pressure ahead of IPOs, or an attempt to block cheaper open-weight Chinese models. Kanter answers that neither the generous nor the cynical reading justifies an antitrust exemption. Companies do not need to coordinate with rivals to ship safe products, he argues, comparing it to Boeing and Airbus: Boeing's door problem wasn't Airbus's fault and didn't require joint slowdown, just better manufacturing. He extends the same logic to AI agents: when an agent, whether human or AI, does something harmful, the company that deployed it should be held liable, similar to an employer answering for an employee's conduct. He does carve out a narrow, legitimate form of industry cooperation: shared threat-intelligence clearinghouses for malicious bots and similar risks, which he says antitrust law does not prohibit. What he says the government should actually do is set enforceable rules, comparable to traffic lights and speed limits for cars, including liability consequences (he floats pulling a product from the market, calling it 'robot jail') for companies that ship unsafe systems. He points to the products liability case against Meta over Facebook and Instagram child safety as evidence the existing legal framework already works, while noting it took a decade to land, which he calls the real problem. He also mentions that the Google and Ticketmaster antitrust cases he brought were won outright, and that the Apple case is still active under the Trump DOJ, which he says has surprised many observers given the administration's largely hands-off posture on AI. The segment also notes an unusual political alignment: libertarian former Trump AI adviser David Sacks has been approvingly retweeting former FTC chair Lina Khan's position that no antitrust exemption is needed. The excerpt provided ends mid-sentence, before Kanter finishes describing what safer products would look like in practice.

Key facts

Jonathan Kanter, former DOJ antitrust chief under Biden and now a law professor at WashU and a technology-policy professor at Carnegie Mellon, argues on The Verge's Decoder that AI labs seeking to coordinate on safety don't need an antitrust exemption.
The host cites researchers at Anthropic and Google DeepMind who have quit publicly over safety concerns, and unnamed researchers who put the chance AI kills everyone at greater than 10 percent.
Kanter compares AI safety obligations to Boeing and Airbus: a company's product defect is its own manufacturing problem, not something requiring joint slowdown with a rival.
He argues companies should be held liable when an agent they employ, human or AI, causes harm, and points to the products liability case against Meta over Facebook and Instagram child safety as proof the legal framework already works, though it took a decade to resolve.
Dario Amodei's phrase for slowing AI development, 'pace the frontier,' and former Trump AI adviser David Sacks retweeting Lina Khan's anti-exemption stance both feature as evidence of an unusual alignment across the debate.

Why it matters

AI labs have been floating antitrust exemptions as part of a broader push to slow development and coordinate on safety, and critics have called this a bid for regulatory capture or a cartel. Kanter, someone who spent years leading US antitrust enforcement against Google, Apple and Ticketmaster, weighing in against the exemption reframes the debate: the argument that safety requires industry-wide coordination, rather than each company simply building safer products, does not hold up under existing law, in his reading.

Who it affects

The debate concerns the major frontier AI labs named in the piece, including OpenAI, Anthropic and Google DeepMind, plus Elon Musk's ventures, as they weigh calls to slow development against IPO pressure and competitive dynamics; the public safety warnings themselves come specifically from researchers at Anthropic and Google DeepMind. It also touches US antitrust policy more broadly, since Kanter's own unresolved case against Apple continues under the Trump DOJ, and any company whose AI agents could cause harm to third parties, since he argues employer liability should apply there as it does for human employees.

How to use it

Treat this as one former regulator's argument, not a settled legal standard: Kanter's position is that existing products liability law, plus government-set rules comparable to traffic laws, are the tools to reach for, rather than pushing for a carve-out that lets competitors coordinate. He does allow one narrow exception, a shared clearinghouse for threat intelligence such as malicious bots, which he says doesn't require antitrust protection because the law doesn't prohibit it.

How solid is it

This is a first-person interview with a named, credentialed source speaking on the record, which is a solid grounding for his own stated opinions and case history. The extinction risk figure ('greater than 10 percent') is attributed only to unnamed 'other researchers,' not sourced or named in this material. The Boeing/Airbus and exploding-car comparisons are Kanter's own illustrative analogies, not reports of real incidents. The provided text is also truncated mid-sentence, so the interview's remaining substance and any concluding views aren't captured here.

Risks and caveats

No date is given for when the interview was recorded or aired. No outcome or regulatory decision is reported: this documents a debate and Kanter's opinions, not any action actually taken on an AI antitrust exemption. Dario Amodei's title at Anthropic isn't stated in the source. The piece is the first half of a two-part series, so context from the second part is missing here.

“these companies do not need to coordinate in order to deliver safe and secure products to society”

— Jonathan Kanter