Anthropic's Amodei says open AI models just shift power to chip owners
A public fight over AI regulation and the concentration of power broke out on X this week, pulling in some of the industry's best-known figures. It started with a claim investor Gavin Baker made on the All-In Podcast: that Anthropic CEO Dario Amodei had told people internally the company could one day be the only private AI firm left in the world, with nothing beside it but governments. Anthropic researcher Sholto Douglas called the account "totally false," saying economic concentration of power is exactly what Anthropic fears most, in a market he described as already the most competitive in the world.
Baker then framed what he saw as the real choice. If AI is potentially dangerous, he argued, there are only two stances: treat it as too dangerous to spread widely and concentrate it among a few companies and politicians, or treat it as too dangerous to concentrate and spread it as far as possible. Betting on some absolute power to benevolently look after humanity, he said, has rarely worked out in history. Former Meta chief researcher Yann LeCun backed the open side, comparing AI to the printing press and the internet: it amplifies human intelligence by widening access to knowledge, and societies need a diversity of AI systems with different values for the same reason they need a diverse press.
Amodei answered in only his fourth X post of the year. He called the regulation-equals-capture framing a false choice, arguing that regulation can also restrain corporate power and help ordinary citizens when good institutions anchor power in ideas rather than individuals. He said Anthropic deliberately designs its proposals to slow down leading labs and favor smaller ones, pointing to California's SB53 law, which fully exempts companies below a revenue or training-cost threshold, as evidence. Open models, he argued, only partly solve the centralization problem: because their weights are freely available, they just shift the concentration of power toward whoever has the most computing power and AI chips, which brings it right back to the big labs. AI centralizes by its very structure, he said, mainly because of scaling laws rather than because of regulation.
David Sacks, until recently the White House AI adviser, answered with a nine-part reply. He noted that Amodei never actually disputed Baker's original claim, even though he easily could have. He called Amodei's regulatory-bubble argument a straw man, since almost no one genuinely believes every regulation amounts to capture, and cited Nobel laureate George Stigler's definition of regulatory capture: an industry acquires the rules and runs them mainly for its own benefit, while the benefit to the public stays scattered and unorganized. Sacks said Anthropic understands this mechanism well, pointing to its hiring of several former AI officials from the Biden administration while it built out a large government-relations operation. He mocked Amodei's proposal for a federal agency to review top models before release as "an approval office for AI," arguing it would create long waiting lines and slow the US down against China, which he said would not adopt similar requirements.
Baker closed by arguing the debate is effectively already settled: almost every major AI company except Anthropic has signed a letter from Nvidia CEO Jensen Huang backing open models.
Key facts
- Investor Gavin Baker said on the All-In Podcast that Amodei had told people internally Anthropic could one day be the only private AI firm left in the world; Anthropic researcher Sholto Douglas called this "totally false."
- In his fourth X post of the year, Amodei argued open-weight AI models do not prevent centralization because power just shifts to whoever controls the most computing power and chips, bringing it back to the big labs.
- Amodei cited California's SB53 law, which fully exempts companies below a revenue or training-cost threshold, as evidence his regulatory proposals are built to slow down big labs rather than protect Anthropic.
- David Sacks, until recently the White House AI adviser, replied in nine parts, invoking economist George Stigler's definition of regulatory capture and calling Amodei's proposed model-review agency "an approval office for AI" that would slow the US down against China.
- Baker said almost every major AI company except Anthropic has signed Nvidia CEO Jensen Huang's letter backing open models.
Why it matters
The exchange puts a public face on a fault line running through the AI industry: whether the technology is safer concentrated among a few well-governed labs, as Amodei argues, or safer spread as widely as possible, as Baker and LeCun argue. Because Amodei has repeatedly called for federal oversight of frontier model releases, how this argument plays out shapes where US AI policy heads next, and who gets to write the rules that follow.
Who it affects
Anthropic and rival labs building frontier models, open-weight model providers whose products are the direct subject of Amodei's critique, investors like Baker who hold positions across the AI market, policymakers weighing proposals such as SB53 or a federal model-review agency, and ultimately anyone affected by whichever regulatory regime the debate helps shape.
How to use it
Treat this as a map of the two camps rather than a settled argument: Amodei's side says unregulated scaling concentrates power in a handful of labs regardless of open weights, while Baker, Sacks and LeCun's side says regulation itself is what locks in the incumbents. Watching concrete moves, such as who signs Jensen Huang's open-model letter or how SB53's thresholds are drawn, is a better gauge of where the industry is actually heading than the rhetoric on either side.
How solid is it
The account rests on a public X thread rather than a formal debate or official filing, so most of it is paraphrased argument rather than direct quotation; the one verbatim quote captured is Sholto Douglas's "totally false." Baker's account of what Amodei said internally is also secondhand, relayed from the All-In Podcast rather than confirmed by Anthropic directly.
Risks and caveats
Both sides have a stake in how the story is read: Baker and Sacks have public records opposing heavier AI regulation, and Amodei runs the company whose proposals are under attack. No neutral arbiter settles the dispute here, and the strongest claims on each side, Baker's account of a private remark and Sacks's characterization of Anthropic's motives, cannot be independently verified from this exchange alone.
“totally false”
— Sholto Douglas, Anthropic researcher