Wanting to Save the World Is No Excuse for a Monopoly Agreement
AI labs want to jointly slow down model development, but four giants sitting down to talk about deceleration is, legally, a textbook antitrust conspiracy — wanting to save the world is not a defense.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Suing a hundred John Does doesn't fly with the judge
Susquehanna spotted someone buying large amounts of short-dated put options in Chinese brokerage stocks before cross-border trading was restricted, so it sued 100 John Does to freeze the accounts. But the judge denied the preliminary injunction: it could not be proven that all 100 of these people were conspiring to trade on inside information. Levine's breakdown is that only a tiny few may have actually gotten inside information, while the rest just saw the tape move and piled in — and piling into publicly visible moves is itself entirely legal. The ones who can actually figure out who is who are the securities regulators, not the market makers who got picked off.
— Matt LevineThe market maker may not have lost that much
Susquehanna claims it lost about $70 million on these puts, but Levine points out that this is only half the story: as an options market maker, after selling puts it would usually hedge immediately, perhaps by shorting the stock, or by buying puts back from another client or another market maker. If Citadel Securities and Susquehanna were both selling and both buying that day, they were effectively betting against each other and should in theory be suing each other. So the "lost $70 million" figure obscures the fact that they were also hedging, and may even have been net up on the day.
— Matt LevineNot training new models is actually more profitable
Levine offers a counterintuitive business logic: model capabilities now run ahead of deployment in many domains, and Anthropic and OpenAI are working with private equity firms to push AI into enterprises. If they stopped model development for a year, they could make a lot of money selling existing models, with quite good unit economics. But competitive pressure forces every frontier lab to keep burning huge sums training the next model — both to win customers on the leaderboards and to win researchers, because researchers want to go to the best lab. So "slowing down" is good for business, but no one dares unilaterally lay down their weapons.
— Matt LevineFour of them sitting down to talk deceleration is collusion
Since no one dares slow down unilaterally, could the four frontier labs sit down and reach an agreement? Levine says this is legally a classic antitrust conspiracy — like the case in which BlackRock was sued by several state attorneys general: ESG investors got coal companies to produce less coal on the grounds that global warming is bad for humanity, but the effect was higher coal prices, making them the OPEC of coal. The line in the ruling is "wanting to save the world is no excuse for an antitrust conspiracy." Dario's memo also added a footnote saying antitrust law would need appropriate amendment before they would be allowed to sit in a room and discuss shutting down killer robots.
— Matt LevineThe founding documents are all handwritten by humans
Sylvia read Kevin Roos's book on the AGI race (out October 6; she had a pre-read copy) and found a contrast: every dramatic turning point in the book corresponds to a human-handwritten memo — Dario's Pacing the Frontier, the Say AGI memo, and so on. Yet what everyone discusses now is "was this written by AI." Reading Dario's memo, she instinctively wondered "was this written by Claude," and concluded it did not read like AI at all — these people are philosophers, writing original founding texts. But the next generation will probably just ask Claude what to do.
— Sylvia Killingsworth401k has no funding gap, only a balance that's too low
A pension is a future cash flow a company has promised, and the company must invest to match that liability; invest wrong and you get a funding gap, which is why ERISA steps in to regulate pension insolvency. A 401k is not a promised cash flow, just a pool of money: if stocks fall, they fall, and there is no such thing on the books as a "billion-dollar funding gap," only "the number in your account is lower than you want." This eliminates concentration risk — with a 401k, pension insolvency can never happen. The cost is that risk shifts from the company to the individual.
— Matt LevinePensions were actually great clients for Wall Street
The popular narrative is that the shift from pensions to 401ks was good for Wall Street, because it turned people who would have retired on a pension into retail clients trading their own accounts. Levine pushes back: a pension is itself an excellent client for Wall Street, because its time horizon is long and it can make the spicy investments retail investors won't touch — private equity, private credit. The 401k, over these decades, became cheap index funds, and retail investors actually pay extremely low fees. So now the financial industry is pushing to "revive the pension," stuffing private assets into retirement accounts, because you can't make any money charging one basis point on a public stock index.
— Matt LevineRetail says long-term, but its hands want liquidity
Private credit BDCs raise money from retail investors on the logic that "you have a 30-year horizon, you don't need liquidity, you should take the liquidity premium." Then this year bad news hit software companies, investors rushed to redeem, and the BDCs could only say that each quarter only 5% of people would be allowed to redeem. Levine calls this a "psychological liquidity mismatch": the fund really is making long-term investments and really should not sell long-term assets to meet redemptions; but retail investors psychologically want more liquidity than the fund offers. In theory, selling private assets to retirement savers is right; in practice, it is very hard.
— Matt LevineIn their own words · checked verbatim
To some people, it feels wrong to buy options. Because you see a big options trade and you have a good guess about why. If you see, looking at public stock tapes, if you just see a public trade and you're like, that must be insider trading, and then you go trade on it, people are like, that's insider trading too, but I disagree. I think that is totally fine.
Matt Levine5:11
That's going to work at the making money factory. That's not going to work at the killing everyone factory.
Matt Levine16:21
And I quote, trying to save the world is no excuse for an antitrust conspiracy.
Matt Levine20:27
And, like, by the way, that has to be wrong. Like, obviously, trying to save the world is a very good excuse for an antitrust conspiracy.
Matt Levine20:27
But that's just like just we're just in like a phase transition right like most of the code most of the algorithms behind modern ai are were invented by humans because who else would have done it, right?
Sylvia Killingsworth23:28
And there's just a misunderstanding, right? Where the private credit funds are like. Is that a liquidity mismatch? It's not. I don't like that. It's a psychological liquidity mismatch.
Matt Levine32:44
And it's also just like a good economic proposition for the financial industry because you can't make money giving people public equity exposure and charging them one basis point.
Matt Levine33:44
Figures
| Number of John Doe defendants sued by Susquehanna | 100 | 2:05 |
| Loss Susquehanna claims on the relevant puts | about $70 million | 8:13 |
| Publication date of Kevin Roos's new book | October 6 | 22:28 |
| Share of private credit BDC redemptions allowed per quarter | 5% | 32:44 |
Glossary
- John Doe
- A defendant in a lawsuit whose identity is unknown, designated John Doe 1 through 100.
- delta hedge
- An options market maker buying or selling the underlying stock after selling options to hedge directional risk.
- BDC
- A listed vehicle that raises money from retail investors and invests in private credit and other non-public assets.
- target date fund
- A fund that automatically shifts from stock index funds to bond index funds as the retirement date approaches.
How to listen
Founders and investors interested in the business logic of the AI race, insider-trading evidence-gathering, and the evolution of the US retirement system.
The opening chit-chat about Absalom, Absalom and reading habits can be skipped.