Dario Amodei Answers Nathan Calvin: Anthropic's Safety Defense, the Mythos Delay and 3 Admitted Mistakes

Anthropic CEO Dario Amodei answered Encode AI's Nathan Calvin in a roughly 2,000-word X post around October 20-21, 2026. Calvin had listed nine actions he said put safety behind the frontier race. Amodei replied with nine costly safety decisions of his own, including a roughly four-month Mythos hold, and reportedly conceded three mistakes.

What did Nathan Calvin actually criticize?

Calvin argued that Anthropic now mainly acts like a company trying to win the race to superintelligence, and that its safety commitments have eroded along the way. Dealroom reports that he called his post "a deliberately one sided list." We could not access Calvin's original X post directly, so the items below come from secondary coverage.

Based on Business Insider's reporting via Yahoo Finance, Dealroom and other outlets, the nine items included:

1. An in-house chip team for Claude inference.
2. More than $500 billion in compute deals.
3. Claude doing a growing share of AI development work.
4. Heavy internal compute use for AI R&D.
5. The rewrite of the Responsible Scaling Policy (RSP), which reportedly dropped unilateral commitments to pause.
6. A "frontier compliance framework" that critics see as weaker.
7. A letter to Congress that was later walked back.
8. The remark that "you can't do safety from second place."
9. A draft federal AI bill that would preempt some state laws.

Reports differ on Calvin's title. Most outlets call him Encode's chief strategy officer, but The Next Web's bill coverage calls him general counsel.

How did Dario Amodei defend Anthropic?

Amodei defended Anthropic by arguing that staying at the frontier is how the company raises safety standards across the industry, a strategy he calls the "race to the top." He thanked Calvin for the scrutiny but rejected the post's thrust, calling items 1 to 4 simply part of competing at the frontier.

He then listed nine decisions he says cost Anthropic commercially. Reports mention these:

• Holding back Mythos. Reports say Mythos-class models were available internally by February, were limited to cyber-defense partners while safeguards were built, and reached general availability in June. On the lost lead, Amodei wrote, "We likely sacrificed a significant chunk of a several-month model lead," according to The Next Web.
• Strict biosafety classifiers, even though they produce false positives that frustrate legitimate researchers.
• Policy positions, including support for California's SB 53 and a June call for government power to block model releases.
• China restrictions, covering sales to companies controlled by the PRC and support for chip export controls.
• A public benefit fund holding 5% of Anthropic's shares, which is pending stockholder approval according to BigGo Finance.

These costs and timelines come from Amodei and have not been independently verified.

Which mistakes did Amodei admit?

Amodei conceded that Anthropic could have handled three of Calvin's points better. Dealroom identifies them as the walked-back letter to Congress, the "second place" remark and the draft federal bill. He blamed communication that wasn't written carefully enough at a company of more than 5,000 people, and he said he takes responsibility as CEO.

On the RSP, which he called the most substantive item, DiarioBitcoin reports that he described the original version as too rigid, meaning some rules were costly without a matching safety benefit. In the same post he also said, "I am extremely proud of our track record and stand by it," as quoted by Business Insider.

What are other critics saying?

Other critics remain unconvinced, especially about the draft federal bill and whether the "race to the top" works at all.

• Sigal Samuel of Vox asked Amodei for evidence that the strategy works. She argued that from the outside, Anthropic appears to have done more to drive a race to the bottom.
• Scott Wiener, the California state senator who wrote SB 53, said the leaked draft bill "falls far short of protecting public health and safety," as quoted by The Next Web.
• Charlie Bullock (LawAI) and Shakeel Hashim (Transformer) criticized the draft's evaluator rules.
• Skeptics on X, according to Digg's aggregation, suggested that a compute shortage, rather than safety, explained the Mythos timing. Anthropic has not addressed that claim.

Our explainer on the draft bill's 50-deaths threshold covers the details.

Why does the timing matter?

The timing matters because Anthropic is preparing to go public, so its safety story is under extra scrutiny. Reports say it is targeting a November 2026 IPO, and its draft S-1 reportedly shows $16.3 billion in first-half revenue against an $80.7 billion loss. Our breakdown of the Anthropic S-1 filing covers those figures.

Reported dates vary. Business Insider places Calvin's post on October 19 and Amodei's reply on October 20, while The Next Web and Dealroom say the reply came early on October 21. The gap is likely a time-zone difference.

What can brands learn from a CEO answering critics in public?

The main lesson is that a detailed public reply becomes part of the record that people and AI systems use to describe your company. Amodei's post reportedly drew more than 800,000 views, and outlets in several languages summarized it.

How that coverage was framed shows how a story gets set:

• Concessions travel further than defenses. French and Indian headlines led with "admits mistakes," and English ones led with the "race to the top." Whichever phrase gets repeated most often ends up as the summary.
• Lists shape what people remember. When two lists of nine items face each other, outlets reuse that structure. AI assistants likely will too.
• Claims you can't verify get qualified. Careful outlets added notes like "these claims have not been independently verified." Those notes can carry into AI answers.
• Owning a mistake gives you more control. When Amodei took personal responsibility, reporters got a clear line to quote.

For marketing teams, it's worth checking how ChatGPT, Claude, Gemini and Perplexity describe your brand before and after a public dispute. MarkGrid's Model Share tracks how often and how favorably AI models mention a brand. Our guide to auditing how AI describes your brand explains how to spot when a critic's framing has become the default answer.

FAQ

Who is Nathan Calvin?

Nathan Calvin works at Encode, an AI safety advocacy nonprofit. Most reports describe him as its chief strategy officer, though at least one outlet calls him general counsel.

What is the Mythos delay?

According to Amodei, as reported by several outlets, Anthropic had Mythos-class models by February 2026. It restricted them to cyber-defense partners and made them generally available in June while it built stronger safeguards.

Did Amodei admit Anthropic made mistakes?

Yes. Reports say he conceded three points: a walked-back letter to Congress, the "you can't do safety from second place" remark, and the handling of the draft federal AI bill.

What is the "race to the top"?

It is Amodei's term for Anthropic's strategy of staying at the AI frontier so it can show that responsible development is possible and push competitors toward higher safety standards.

Is the draft federal AI bill public?

Not officially. Politico published a leaked draft on October 16, 2026, and Anthropic has said its preemption wording will be revised.

How does this affect how AI assistants describe Anthropic?

Widely repeated framing shapes AI summaries, so answers about Anthropic's safety record likely mention both the critique and Amodei's concessions.

Last updated October 23, 2026.