Amodei defends ‘race to the top’ against Calvin’s critique
Confirmed
The takeaway
On Oct 21, 2026 (Oct 20 US time) Dario Amodei posted a long reply on X to Encode’s Nathan Calvin, who had argued that Anthropic’s actions over the past year look like a company trying mainly to win the race to superintelligence.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 21 October 2026
Your AI and this story
- GPT-6 Astra174 days after its cutoff
- Claude Opus 5.5113 days after its cutoff
- Gemini 4 Argon204 days after its cutoff
- Grok 4.7143 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 113 days before it.
Key facts
- Calvin’s post (Oct 19, ~399k views) listed nine actions: an in-house chip team; >$500B in compute deals; Claude ‘leading’ 26% of AI-development tasks by August (0% in March); heavy internal AI R&D compute; RSP v3 dropping unilateral commitments to slow or stop; a ‘frontier compliance framework’ with no binding rules on internal deployment; a letter to Congress calling Claude’s hacking incidents a ‘misconfiguration’; ‘you can’t do safety from second place’; and the draft bill pre-empting state law
- Amodei’s reply (Oct 21 02:26 UTC, ~664k views) lists: (1) role in SB 53, arguing for a stronger version; (2) strict bio classifiers despite false positives, after ‘concerning results in bio uplift trials’ and blocked bioweapon attempts; (3) holding Mythos-class models (first had in February) back until June for cyber safeguards, ‘likely’ sacrificing ‘a significant chunk of a several-month model lead’; (4) calling in June for government power to block model releases; (5) the events of February 2026; (6) chip export-control advocacy; (7) being the first and only frontier company to restrict sales to PRC-controlled companies; (8) reports and model cards; (9) the 5% public benefit fund
- On Calvin’s items 1–4: ‘simply examples of what it means to compete at the frontier, and we stand by them’. He added that Anthropic is ‘nearing’ the point where AI-driven acceleration gets ‘fast enough that it is hard for safety to keep up’, which drives its thinking on ‘pacing’
- On items 7–9: they fall under ‘rhetoric or policy communication that wasn’t carefully written or resulting from fast-changing circumstances’; ‘Anthropic is over 5,000 people’; ‘I am of course the CEO, and I take responsibility’
- On RSP v3 (item 5): the old RSP created effortful requirements that were not very important for safety; the 2023 RSP was described as ‘an early iteration’. On the FCF (item 6): ‘I promise to look into it further’
- Closing: ‘I will have more to say soon about how we are going to further live our values in this moment of rapid change’
- Coverage: Techmeme (‘admits mishandling a draft AI bill’), Business Insider, Fox News (‘Anthropic CEO defends record on safety, sales to China’)
What happened
The exchange grew out of criticism of Anthropic’s leaked draft federal AI bill. Calvin, a lawyer at Encode, wrote a long post asking “what is Anthropic?” and argued that its recent actions fit a company racing to superintelligence with safety second. Amodei answered in a ~11,000-character post: he defended the strategy, listed costly decisions and accepted fault on communications. Both posts were read in full through api.fxtwitter.com.
Why it matters
It is the most detailed public defence of Anthropic’s strategy by its CEO in the run-up to the IPO. It also contains new first-hand claims: Mythos-class models existed in February, Anthropic saw concerning bio-uplift results, and it is “nearing” the point where safety struggles to keep pace with AI-driven acceleration. He promised further announcements.
Sources
2 sources from 1 site. Numbers match the chips in the text.
2 sources: 1 primary, 1 reaction
Primary
- Dario Amodei on X: reply to Nathan Calvinx.com, official
Reactions
- Nathan Calvin on X: ‘what is Anthropic?’x.com, discussion
Changes
- Filed (Techmeme)