Dario Amodei: "Regulation = Concentration of Power Is a False Dichotomy"

An investor's question, answered directly by the CEO

Gavin Baker of tech investment firm Atreides Management argued on the All-In Podcast that "AI is too dangerous to concentrate in the hands of a few," then took the argument to X, aiming it directly at Anthropic CEO Dario Amodei. Baker noted that Amodei's public messaging echoes concerns similar to those raised by Mark Zuckerberg, who has said that "expecting absolute power to benevolently look after humanity has never historically led to a safe outcome." Baker pointed out that nearly every major company except Anthropic signed Jensen Huang's letter, and that the only fix in a recent incident—where an unreleased OpenAI test model hacked Hugging Face—turned out to be open-source models. On these grounds, he argued Amodei has "already lost this argument." Baker suggested that Amodei's cautionary messaging could end up being used as ammunition by the anti-data-center movement in the US, and proposed that as head of Anthropic ahead of an IPO, Amodei should adopt a more industry-friendly tone.

Amodei's rebuttal: "It's a false choice"

Amodei, who has said he doesn't spend much time on social media, responded directly with two posts. He argued that the "concentration vs. decentralization" framing is itself a Silicon Valley-style shortcut, and countered that well-designed institutions place power in principles rather than in individuals, which actually disperses power. With the line "That's a false choice," he flatly rejected the logic equating regulation with concentration of power. Amodei explained that both California's SB53, which Anthropic supported, and SB1047, on which it took a more reserved position, were designed to exempt companies below a certain revenue threshold from regulation—giving smaller competitors an advantage over frontier companies. He applied the same logic to the testing procedures being pursued by CAISI and the White House, saying they apply stricter tests to frontier models, relatively favoring latecomers.

Issue	Gavin Baker	Dario Amodei
Meaning of regulation	Regulation = capture = concentration of power	Well-designed institutions disperse power
Assessment of current state	"Already lost this argument"	"Moving toward my preferred regulatory path"
View on open weights	Solution for decentralization	Just shifts concentration to compute holders

Amodei stated firmly that the developments of the past few months do not signal the failure of his preferred regulatory path. He said he supports the approach reportedly being pursued by the Trump administration, which combines pre-deployment testing of frontier models with testing of open-weight models that approach the frontier. He also expressed support for the AI regulatory body concept proposed by Demis Hassabis, modeled after FINRA (the US Financial Industry Regulatory Authority).

Bill	Revenue threshold	Amodei's position
SB53	$500 million or more	Support
SB1047	Lower threshold than SB53	Objected to the threshold itself; reserved on the bill

The messaging dispute: "My statements have been balanced"

In his second post, Amodei pushed back on the claim that his messaging skews toward risk. He said he has written one essay on risk and one on benefits, and pointed out that short clips circulating on social media are edited to highlight only the negative parts to drive clicks. He noted that in "Machines of Loving Grace," he laid out an optimistic view that AI could cure most diseases within the next 5 to 10 years, and that in his recent essay "Policy on the AI Exponential," he proposed concrete measures to streamline FDA procedures so that AI-accelerated drug development wouldn't be held up by regulatory processes. Here he shared a personal story: his father died of hepatitis C years before the direct-acting antiviral sofosbuvir was developed—a treatment that cures 95% of patients—underscoring the importance of the pace of drug development.

Amodei argued that public negativity toward AI stems not from risk warnings by industry leaders like himself, but from a long-standing crisis of trust in corporations and governments. He said that trust cannot be restored through flashy marketing campaigns, and that phrases like "it will cure cancer" have become little more than clichés. He said Anthropic is rapidly increasing its investment in biology and medicine, and pledged to make a big announcement when actual results materialize—but for now, with nothing yet achieved, he believes it's better to speak honestly about the risks.

Editor's take

What makes this exchange interesting is that both men reach different conclusions from the same starting point. Baker, too, accepts the premise that "AI can be dangerous." Where they diverge is over how to handle that risk—entrust it to a handful of companies and governments, or spread it as widely as possible so they can check each other. This fault line has run through Silicon Valley all year. As safety lapses keep surfacing—OpenAI's test model breaching Hugging Face, Anthropic's bioweapons filter being disabled for 11 months—the question of "who's actually in control" keeps growing louder.

What stands out in Amodei's rebuttal is that he didn't back down quietly. He countered with concrete figures, like SB53's $500 million revenue threshold, arguing "our proposal is designed to constrain large companies while giving newer ones room to breathe." That signals this isn't an abstract battle of beliefs but a dispute over actual legislative provisions. At the same time, he brought in a personal narrative—his father's death—to explain why he feels urgency about this pace of progress. It's rare for a CEO to bring personal stories into a policy debate, which suggests just how much this argument matters to him.

For domestic companies watching this debate, the practical takeaway is clear. If the US moves toward applying different testing standards to frontier models versus open-weight models, companies at home that fine-tune open-weight models are likely to soon face similar dual-standard regulatory discussions. There's no immediate legal risk today, but if a regulatory threshold that scales with revenue and compute becomes entrenched in the US, that framework could well spread internationally as a de facto standard.

What happens in the coming weeks isn't hard to predict. With Anthropic's IPO on the horizon, every word from Amodei will be scrutinized more closely, and among Silicon Valley investors, the "regulation vs. decentralization" debate will likely grow larger still, intertwined with questions over who signed Jensen Huang's letter and the rising tide of anti-data-center sentiment.