Why Dario Amodei Went Line by Line Against an AI Safety Critic
Encode AI's Nathan Calvin says Anthropic now acts like a company racing for superintelligence first. Dario Amodei answered with nine costly safety calls and three narrow concessions, but sidestepped the core charge.
Most CEOs heading toward a public listing let the comms team handle their critics. Dario Amodei answered his own. Early Wednesday, the Anthropic chief executive posted a long, point by point reply on X to Nathan Calvin, chief strategy officer at the advocacy group Encode AI, who had argued that Anthropic now behaves less like a safety lab and more like a company determined to win the race to superintelligence. According to The Next Web, the reply drew more than 800,000 views.
The exchange puts Anthropic's founding bet on trial in public. Amodei's theory is that a responsible lab at the frontier drags everyone upward. Calvin's question is sharper. What if being at the frontier is the thing that bends the lab?
The charge sheet
In an October 19 post, Calvin laid out nine actions from the past year that he said looked "most consistent" with a company "primarily trying to win the race to super intelligence." His list included more than $500 billion in compute deals, an in-house chip team, rising use of Claude to lead internal AI development work, the switch to version 3 of the Responsible Scaling Policy, a later walked back letter to Congress, the line "you can't do safety from second place," and a draft federal bill he said would preempt significant parts of state law.
He called the list "deliberately one sided" and credited Anthropic's safety research. His sharpest point was about the pattern.
"When Anthropic strays and makes mistakes, why do the mistakes systematically tend to guard and favor Anthropic's interests?" Nathan Calvin, Encode AI
Amodei's counter list
Amodei called Calvin's comments constructive and honest, then said he disagreed "with the thrust" of the post. He answered with a rival list of nine decisions he described as commercially costly and good for safety. Among them:
- Mythos on a leash. Amodei wrote that Anthropic first had Mythos class models in February but held general availability until June, when stronger cyber safeguards were ready. "We likely sacrificed a significant chunk of a several-month model lead," he wrote.
- Bio classifiers. Amodei said Anthropic's biological safeguards are strict enough that "biologists joke about how restrictive our classifiers are,".
- Policy fights. He cited SB 53 in California, his June call for government power to block model releases, and chip export controls on China.
- Money on the table. Amodei said Anthropic is, as far as he knows, the only frontier company restricting AI sales to PRC controlled firms, and that it just pledged 5% of the company to a public benefit fund.
"I think that even if I conceded every item in your list, I would happily hold up mine next to it." Dario Amodei, Anthropic
What he gave up, and what he didn't
Amodei agreed Anthropic could have handled the Congress letter, the "second place" remark and the draft bill better, blaming communication that "wasn't carefully written" inside a company of more than 5,000 people. "I am of course the CEO, and I take responsibility," he wrote.
On the four items about compute, chips and internal R&D, he gave no ground, calling them "simply examples of what it means to compete at the frontier." He described the Responsible Scaling Policy change as "the most substantive item" but defended it, arguing the old version was too rigid. On the separate frontier compliance framework, he promised to "look into it further."
That split is the tell. Amodei conceded the words and defended the machinery. Calvin never claimed Anthropic does nothing for safety. He argued it pushes the capability frontier far harder than the safety frontier. Amodei mostly answered a different question, whether Anthropic pays any price for safety at all.
Amodei did leave one door open. He wrote that Anthropic is "nearing the point" where AI acceleration makes it hard for safety to keep up, that this drives internal thinking about "pacing," and that he will "have more to say soon." For a company whose valuation draws intense investor scrutiny, talk of slowing down is striking. Whether it becomes more than a promise is what Calvin will be watching.
