Dario Amodei used two posts on X on August 15 to argue that AI policy does not force a choice between concentrating power in a handful of companies and spreading it widely. Responding directly to investor Gavin Baker, the Anthropic chief executive said critics who reduce all regulation to regulatory capture “often underrate the decentralizing power of objective and fair institutional processes.” His illustration was a court system that can feel stuffy and elitist but still protects individuals better than what he called mob justice.
Amodei got specific about what that looks like in practice. He said Anthropic’s own policy proposals are built to disadvantage frontier labs while advantaging smaller competitors. California’s SB 53, which Anthropic supported, exempts companies below roughly $500 million in revenue or training costs. SB 1047 carried a lower threshold, which Amodei said Anthropic argued should have been set higher. The testing regime Anthropic has pushed at CAISI and the White House applies more rigorous checks to frontier models than to those trailing behind, and the “Pacing the Frontier” letter he cites would slow only the very best models rather than the ones still catching up. Each mechanism draws its line using revenue, training cost, or capability rather than a regulator’s discretion over which company wins. The thread does not say who would administer or audit those thresholds once written into law, only that Anthropic has spent months advocating for them.
The second half of the thread carries more weight than the regulatory argument. Amodei rejected the idea that his own warnings about AI risk are what turned public opinion negative. He called the backlash “fundamentally a crisis of trust,” writing that ordinary people “don’t trust companies, governments, or the tech industry” and assume every institution is “cooking up some new way to screw them over,” a suspicion he says predates AI by decades.
That is where Amodei conceded ground that frontier lab executives rarely concede in public. He wrote that the most accurate criticism of Anthropic and its peers is not their messaging but that they “haven’t yet delivered on our big promises to benefit the world.” No marketing campaign fixes that, he added: the actual fix is “actually curing cancer,” not describing it. Coming from the head of a company that has built its brand partly on the promise of curing disease, that is a real admission rather than a rhetorical one.
Amodei is not a neutral commentator on any of this. Anthropic is reportedly preparing for an IPO that could value the company near $2 trillion, a deal AI Insiders covered on August 15. The revenue and compute thresholds he cites as evidence of fairness are the same thresholds Anthropic has lobbied regulators to adopt, which means the executive proposing to constrain frontier labs also gets a hand in defining what counts as frontier. That does not make the argument wrong. It does mean the line separating challenger from incumbent has, so far, mostly been drawn by an incumbent.
For operators tracking Washington, the near-term test is whether the Trump administration’s pre-deployment testing framework, which Amodei says he supports, actually adopts tiered thresholds like the ones Anthropic has proposed. Real thresholds would give smaller labs breathing room. Thresholds set low enough to cover today’s challengers would fold Amodei’s own distinction back into the regulatory capture his critics already fear.
Dario Amodei, in a thread on X posted August 15, 2026, responding to investor Gavin Baker.
