Dario Amodei says safety has cost Anthropic money, and he has a list to prove it.
In a long post on X, the Anthropic chief executive defended the company's "race to the top" strategy. He was replying to Nathan Calvin, chief strategy officer at the advocacy group Encode AI. Amodei listed nine decisions that he said hurt Anthropic commercially but made AI safer. He also said the company held Mythos back for four months for safety reasons.
"Race to the top" is Anthropic's answer to a familiar objection: if you think AI is dangerous, why build it? The company's pitch is that competing beats abstaining, because a lab that sets a higher safety bar can pressure rivals to match it. That pitch only works if the bar is real, and a safety choice that costs nothing proves little. Hence the emphasis on lost business.
The weakness is that the ledger is Amodei's own. A delay is easy to announce and hard to audit from outside, since only the company knows what it gave up and what it would have shipped otherwise. Nine decisions and four months are claims, not measurements.
The exchange also shows who is asking. Calvin runs strategy at an advocacy group, which suggests Anthropic's safety record is being tested by the policy community as well as by competitors.
The better test of the strategy sits outside Anthropic. A four-month hold changes the industry only if rivals feel they must match it, and a delay that rivals simply outrun is a cost with no lever attached.