Anthropic CEO says AI centralizes by nature and open

What happened

An open fight over AI regulation and the concentration of power has broken out on X. Investor Gavin Baker, former White House adviser David Sacks, and Meta researcher Yann LeCun accuse Anthropic CEO Dario Amodei of using fear rhetoric to buy himself a regulatory advantage. It started with a claim on the All-In Podcast.

That's why Anthropic deliberately designs its proposals to slow down leading labs and favor smaller ones, he said. As proof, Amodei pointed to California's SB53 law, which fully exempts companies below a revenue or training-cost threshold from its requirements. Open models, whose weights are freely available, only partly solve the problem, he argued.

They just shift the concentration toward whoever has the most computing power and the most AI chips, which brings it right back to the big labs. AI is centralizing by its very structure, he said, mainly because of the scaling laws and not because of regulation. Investor David Sacks, until recently the White House AI adviser, hit back at Amodei's account in a nine-part reply.

What stands out, he said, is that Amodei doesn't actually dispute Baker's report, even though he easily could. Amodei's regulatory-bubble argument is a straw man, Sacks argued, since almost no one really thinks every regulation is capture.

He pointed to Nobel laureate George Stigler's definition, under which an industry acquires the rules and runs them mainly for its own benefit, while the benefit to the broader public stays scattered and unorganized. Anthropic understands this mechanism perfectly, Sacks said, and has hired several former AI officials from the Biden administration while building out a large government apparatus.

Why it matters

Investor Gavin Baker said Anthropic CEO Dario Amodei had told people internally that the company could one day be the only private firm left in the world, with nothing beside it but governments. " Economic concentration of power, he said, is exactly what Anthropic fears most.

The AI market is already the most competitive in the world, he argued, because every big company is racing to build smarter and cheaper models. Baker then spelled out the heart of the dispute: If AI is potentially dangerous, there are two basic stances.

Either you consider it too dangerous to spread widely and concentrate it among a few companies and politicians, or you consider it too dangerous to concentrate and spread it as far as possible. In Baker's view, betting that some absolute power will benevolently take care of humanity has rarely ended well throughout history. Former Meta chief researcher Yann LeCun also came down firmly on the side of open AI.

Like the printing press and the internet, he said, AI amplifies human intelligence by improving access to knowledge. Societies need a wide diversity of AI systems with different values, for the same reason they need a diverse press. In the end, one person's good AI stands against another's bad AI. The long thread eventually drew in Amodei himself.

In only his fourth X post of the year, Amodei called this framing a false choice. In parts of Silicon Valley, he said, the going assumption is that regulation automatically means regulatory capture, meaning an industry writes its own rules to its own benefit. That's too simple. Regulation can also rein in corporate power and help ordinary citizens, he argued, when good institutions anchor power in ideas rather than individuals.

What to watch

He mocked Amodei's proposal for a federal agency to review top models before release, calling it an approval office for AI. A review process like that would create long waiting lines and slow the US down against China, which won't adopt such requirements, Sacks said. Baker, for his part, holds that the debate is already over: Almost every major company except Anthropic has signed Jensen Huang's letter backing open models.