The Regulatory Moat: When the Frontier Becomes a Cartel

I do not usually break my Friday publishing cadence, but the mask just slipped on the AI industry's endgame.

If you have been watching tech Twitter this week, you saw the explosion. A researcher named Jacob Coxin resigned from Anthropic, dropping a tweet that racked up over 170 million views. His claim was blunt: both OpenAI and Anthropic are acting irresponsibly, racing toward superintelligence, and gambling with human lives. It was the largest viral moment in the history of AI safety.

The ensuing panic was immediate, but the real story isn't the resignation itself. The real story is how the major centralized labs immediately tried to capitalize on it.

Right on cue, Anthropic CEO Dario Amodei published a manifesto on "pacing the frontier." His grand solution? Every frontier AI lab needs to submit to "embedded third-party evaluators" to grade their safety homework. He specifically named an organization called METR as the gold standard for this independent auditing. Sam Altman and Elon Musk immediately chimed in to agree.

If you just read the headlines, it looks like responsible self-regulation. But if you map out the actual network, as ThePrimeTime did brilliantly in a recent breakdown, the illusion of independent evaluation falls apart completely.

The Cabal Layer

To understand what is actually happening, you have to look at the money and the marriages.

METR was spun out of an organisation called ARC, founded by Paul Christiano. Paul happens to be the trustee of the Long-Term Benefit Trust that spun out of Anthropic. He is also married to Ajeya Cotra, who works at METR. Dario Amodei and Paul Christiano used to live together. Holden Karnofsky, an early investor in Anthropic, is married to Daniela Amodei, Dario's sister and co-founder. The funding for all of these interconnected Effective Altruism (EA) groups, including Open Philanthropy, Survival and Flourishing Fund, and METR itself, flows through the exact same tight-knit circle of Silicon Valley insiders, including Dustin Moskovitz and, previously, Sam Bankman-Fried.

Even Jacob Coxin, the supposedly independent whistleblower whose tweet kicked this all off, was revealed by the Wall Street Journal to have quietly coordinated his resignation with anti-AI policy groups beforehand.

METR isn't an independent watchdog. It is the exact same social graph.

David Sacks called this out perfectly: this is an attempt to suspend antitrust norms to form a cartel. Dario Amodei wants to shepherd in superintelligence, but more importantly, he wants the power to tell everyone else they aren't allowed to build without his explicit approval.

The Regulatory Moat

In previous articles where I discussed how you can cut your AI costs with smart model routing, the underlying assumption was that raw model capability is rapidly commoditising. Open-weight models are destroying the margins of closed APIs, and the technical battleground has moved up the stack into compute, context, and workflow orchestration.

But I missed a crucial layer. When your technical moat evaporates, your last line of defence is a regulatory moat.

If you cannot beat open-source efficiency on merit, you make it illegal or impossibly bureaucratic to compete. You convince the government that AI is a uniquely dangerous weapon that can only be handled by a few responsible mega-corporations. And then, conveniently, you set up an independent evaluation board staffed entirely by your former roommates, investors, and siblings to serve as the exclusive gatekeepers.

ThePrimeTime joked that Sam Altman and Elon Musk are only agreeing with Dario to "gas him up" - encouraging Anthropic to slow down and police itself while OpenAI and xAI quietly race ahead. Whether it is a stall tactic or a genuine cartel play, the underlying engineering reality is identical. The centralised labs are trying to pull up the ladder behind them.