AI Giants Want an Antitrust Exemption: Safety or Cartel?
Former DOJ antitrust chief Jonathan Cantor says building safe products doesn't require coordinating with rivals — just as Boeing doesn't need to consult Airbus when a door falls off — so no form of antitrust exemption should be granted.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Building safe products doesn't require coordinating with rivals
Cantor splits AI safety into two questions: what companies should do, and what government should do. His core rebuttal is that companies don't need to coordinate to deliver safe products — Boeing's door falling off isn't Airbus's fault, and doesn't require slowing innovation together with Airbus; a car exploding isn't another carmaker's fault. Likewise, if your AI agent breaks into someone else's system, that's no different from sending an employee to do it, and the employer should be held liable. The only things that truly need collaboration are matters like threat intelligence sharing, such as building a shared library of malicious bots, and antitrust law already doesn't prohibit that kind of collaboration. Only when two companies say "we're competing too hard, we need to slow down together" does it actually touch antitrust law.
— Jonathan CantorEven the most charitable reading doesn't justify an exemption
Cantor offers two readings, from most charitable to most cynical, and both lead to the conclusion that no exemption should be granted. Most charitable: they genuinely fear the pace of innovation is out of control, there are no lines on the road, and they're afraid of driving off the road or hitting someone, so they ask the government to step in — he believes they believe they might destroy humanity, but doesn't think that doomsday scenario is accurate. Most cynical: they've burned so much money they can't sustain it, but pulling back would hurt their valuations and IPOs, so they want someone to step forward and give everyone a "slow down together" permission slip, so they can figure out the economic model before the election. He states clearly that neither reading should earn any form of antitrust exemption.
— Jonathan CantorThe products liability framework has already worked once
The host asks: if I direct my Meta agent to attack you, is that Meta's fault? Cantor says both — companies hire people and build technology, and when people or technology go steal or break into someone else's property on the employer's behalf, liability exists, and it can be products liability or go beyond it. He cites Meta's case over child safety and mental health on Facebook and Instagram as an example, saying the products liability framework has indeed worked recently. But the host immediately points out: it took ten years to work, and that's the problem. Cantor acknowledges this, so government must step in, and the pace of innovation is no excuse for not building safe products.
— Jonathan CantorNow is the time to change liability from "maybe" to "certain"
The host proposes: Meta's products liability case ended in a settlement, Meta made concessions to regulators, but it took a long time. If today we said OpenAI is likely liable for ChatGPT's behavior and Anthropic is likely liable for Claude's, shouldn't we now tighten the screws and legislate clearly that you are definitely liable? Cantor says this is exactly the clarifying work Congress should do, but today's environment is different from fifteen or twenty years ago: back then Section 230 gave online companies broad exemptions, plus a laissez-faire attitude starting in the Obama era, which let tech companies camp out directly in the White House and Congress and tell the government how to write the rules. He says that outcome has destroyed public trust in these companies, which is why this conversation is happening much earlier — "fool me once, shame on you; fool me twice, shame on me."
— Jonathan CantorLeft and right are realigning on antitrust
The host points out a political realignment that confuses him: Trump's AI czar David Sacks retweeted former FTC chair Lina Khan, saying existing antitrust law is sufficient and no exemption is needed; hours later Bernie Sanders and Steve Bannon shared a stage in DC calling for a halt to AI development. Cantor calls this a "realignment of the realignment": during Trump's first term, left, right, and center all worried about Big Tech power, and Lina and he took office with a popular mandate to enforce aggressively; the second term first turned friendly, and now the cracks are reappearing. He adds that this misalignment has long existed — when he testified before Congress he had support from Mike Lee, Chuck Grassley, and Josh Hawley at the same time, and Matt Gaetz even asked him at an oversight hearing, "Have you won enough?"
— Jonathan CantorThe China threat is a convenient shield
The host asks: these companies say America must win, otherwise the Chinese government will control AGI — is this a convenient boogeyman? Cantor says "to a large extent, yes." He goes back to the AT&T breakup: back then AT&T told the Reagan administration that winning the Cold War required a strong national telecommunications network, and the Reagan administration rejected that argument. He says the right way to compete with China is to play to America's strengths, and China isn't doing what we're doing — China restricts children's access to technology, censors domestic information, and concentrates control over companies through state-owned enterprises and affected businesses. So the claim that "the only way to compete with China is to let us ship unsafe products" is absurd. He also says the lack of rules and accountability in the social media era actually led to a lack of innovation in "how to build products safer from the start."
— Jonathan CantorTolerating domestic monopoly for foreign competition is wrong
Cantor's framework is: promote fierce, dynamic competition within the United States, including foreign companies, because America is a market economy, and the chance of success incentivizes people to try new things, invest capital, and make good products. He says the idea that "we need domestic monopolies to compete abroad" is fundamentally opposed to the American way of life — America's founding itself was a revolt against British monopoly over the necessities of life. He acknowledges there are legitimate questions, such as how to respond to foreign below-cost dumping, which this administration is raising; he mentions an 1980s case he read with law students: Japanese companies were accused of cartelizing TV prices in Japan, then cross-subsidizing below-cost pricing in the US to crush American manufacturers, and at the time the Supreme Court used an absurd set of economics to find this didn't constitute an antitrust violation. But he says the way to handle this is not to allow the domestic market to be monopolized.
— Jonathan CantorNVIDIA's circular investments should be scrutinized
The host asks on behalf of the most skeptical listeners: NVIDIA has circularly invested in all these companies, they're effectively already coordinated; the products liability theory will fall flat because OpenAI will never sue Hugging Face, and NVIDIA just bought Hugging Face and is also OpenAI's largest investor — shouldn't all this be broken up? Cantor says "someone should be looking at these questions," which is exactly what they started doing in office: draw some lines, you can't buy whichever company you want and invest in whichever company you want, competition has to be competition, and within the boundaries you can compete as much as you like. He criticizes this administration for allowing a few companies to reach dominant, oligopolistic, or monopolistic positions in order to "beat China," but he asks "what does beating China even mean" — if China has advanced bots invading American companies and government servers, that's a national security, geopolitical, international relations problem, and shouldn't be solved by domestic monopoly, especially when these monopolists are loyal to shareholders rather than the public interest.
— Jonathan CantorAntitrust enforcement won cases but lost on remedies
Cantor sums up his tenure: they brought back aggressive enforcement, using theories that didn't just look at whether consumer prices rose or fell, which was new at the time. They won against Google twice, the Apple case survived a motion to dismiss, the Ticketmaster case even won jury support when the Trump administration tried to settle, and they brought cases on algorithmic pricing and healthcare. He acknowledges the remedy outcome in the Google case was disappointing, but that wasn't the part he presided over — he presided over the wins, then passed the baton to the next administration. Some people, especially in tech policy circles, say this whole thing was a failure that brought no structural change; Cantor says "absolutely not": unfinished doesn't mean failed, and it's inherently harder to act after the market has already hardened.
— Jonathan CantorWhat needs protecting is the next inflection point
Cantor says he has always liked to say: what you really want to do is protect the next inflection point. Especially in technology and platforms, what comes next is likely to be highly disruptive, and competition is most promising at that moment — and it's also the moment when incumbents most aggressively block new players from competing. The host picks up: when you were on the show last time talking about the Apple case you said the same thing, before everything changes, monopolists are most aggressive; Apple was the App Store, Google was bundling products in various ways. Now we're talking about challengers: Johnny Ive went to OpenAI to do hardware, and Apple is suing them. Cantor says yesterday's challengers are often today's monopolists, and he's not against challengers — that's exactly why he opposes antitrust exemptions; he wants challengers to keep charging hard, and he also wants thriving domestic open-source or open-weight models to compete.
— Jonathan CantorPrescription: first clarify the cost of bad outcomes
Asked how he would prioritize if he were a policymaker, Cantor says start with "clarifying the consequences of bad things": if you build a dangerous product, if the door flies off in midair, if your AI breaks into someone else's company, you should be held liable, and you have an obligation to build products that won't do that — this is an extension of the products liability-style framework, and a good starting point. Second, Congress should step in and enumerate the things society values: mental health, competition, information truthfulness, content owners' intellectual property and copyright, and embed these values into the lines on the road, stop signs, and traffic lights, so everyone knows from the start what's allowed and what isn't. He says maybe the next Congress can do it, and he hopes so.
— Jonathan CantorIn their own words · checked verbatim
So the state of the world right now is it's like we've invented cars and trucks, but we have no lines on the road, no traffic lights, and no stop signs and no speed limits.
Jonathan Cantor3:04
So let me give you my interpretation from the most generous to the most cynical, neither of which, TLDR, should result in any sort of antitrust exemption.
Jonathan Cantor5:18
These companies do not need to coordinate in order to deliver safe and secure products to society.
Jonathan Cantor6:18
If you believe your product is going to destroy humanity, then don't build it.
Jonathan Cantor11:35
It's largely a convenient boogeyman.
Jonathan Cantor21:52
The idea that we need monopolies at home in order to compete abroad is antithetical to our way of life.
Jonathan Cantor24:57
China is a country of national champions. The United States is not a country of national champions.
Jonathan Cantor31:09
And I've always been fond of saying that what you really want to do is you want to protect the next inflection point.
Jonathan Cantor39:29
Figures
| AI researchers' estimate of the probability AI exterminates humanity | greater than 10% | 1:00 |
| Time taken by the Meta products liability case | ten years | 10:32 |
| Number of wins against Google during Cantor's tenure | twice | 38:28 |
Glossary
- antitrust exemption
- A legal exception allowing competitors to coordinate actions without being found in violation of the law.
- regulatory capture
- When regulated companies come to dominate the regulatory rules, making the rules serve themselves.
- products liability
- Legal liability of a manufacturer for harm caused by product design or defects.
- Section 230
- US law giving online platforms broad immunity from liability for user content.
- national champions
- A few leading enterprises supported by the government to represent the nation in international competition.
How to listen
Founders, investors, and policy researchers watching AI regulatory trends, antitrust enforcement, and US-China tech competition policy.
The sponsor read at the start, 0:00-0:30, and the production credits after 43:43 at the end.