Anthropic CEO Dario Amodei defended the company's safety record in a nearly 2,000-word post on X on Tuesday, answering a critique from Nathan Calvin, chief strategy officer at the AI-safety nonprofit Encode AI. Amodei said he was a fan of Calvin's investigative work but disagreed with many of his criticisms, and he argued that competing at the frontier does not conflict with Anthropic's safety mission.
Calvin's post, published Monday and running more than 1,500 words, listed nine Anthropic actions from the past year that he said undermined its safety commitments. They included signing more than $500 billion in compute deals and relying increasingly on Claude to help develop Anthropic's own models.
Calvin also pointed to a February 2026 revision of Anthropic's safety policy. The earlier version committed the company to pausing training or releases if safety measures fell behind model capabilities; the revised policy dropped the requirement to independently halt R&D at higher-risk stages, with Anthropic citing the risk that competitors would take advantage.
Key facts
- 01Dario Amodei posted a nearly 2,000-word reply on X to Encode AI chief strategy officer Nathan Calvin.
- 02Calvin's 1,500-plus-word post listed nine Anthropic actions, including more than $500B in compute deals, that he said undercut its safety commitments.
- 03Amodei countered with nine commercially costly decisions, including restricting AI sales to PRC-controlled companies.
- 04Anthropic's draft federal bill would let the government seek to halt a model posing risk of more than 50 deaths or $1B in damage.
- 05Anthropic says it will propose R&D pacing policy options by January 15, 2027, or explain why it proposes none.
Amodei responded with a nine-item list of his own: decisions he said cost Anthropic commercially but carried safety benefits. They included releasing the Mythos model only to a select group of cybersecurity partners, restricting sales of Anthropic's AI services to Chinese state-controlled companies, committing 5% of the company to a public benefit fund, and running biology safeguards that deliberately accept a high rate of false positives to block bioweapon misuse.
“we were the first, and still only, frontier company to restrict the sale of AI services to PRC-controlled companies”— Dario Amodei, Anthropic CEO
Amodei framed the approach as the company's long-running "race to the top" strategy, arguing that showing the technology can be built responsibly raises the bar for the rest of the industry. He said staying at the frontier gives Anthropic more leverage to push for stricter industry-wide rules, and that the updated scaling policy increases transparency and external review.
He also conceded that Anthropic could have handled some issues better, including a draft bill it circulated to Congress this month. The draft would let the federal government seek to halt the release of a model that poses an imminent risk of more than 50 deaths or serious injuries, or more than $1 billion in damage, from threats such as biological weapons, cyberattacks, or loss of control, with the Commerce Department bringing the case before a federal judge.
AI-safety advocates have questioned the draft. Critics said the threshold would make it harder for the federal government to block dangerous models, and some provisions may also limit state governments' authority to regulate AI.
Amodei acknowledged communication mistakes inside what is now a 5,000-person company, while standing by the overall record.
“We'll always try to do better but, again, I am extremely proud of our track record and stand by it”— Dario Amodei, Anthropic CEO
Anthropic published a separate safety-roadmap update on October 19 saying its existing policies mainly target the deployment of dangerous models and that limiting the pace of R&D may become necessary. The company said it will propose related policy options by January 15, 2027, or explain why it is proposing none.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
