The small company that was so afraid of AI in 2023 that its employees couldn’t sleep has today become the largest IPO in history. On the eve of Claude-maker Anthropic’s stock market debut, renowned tech journalist Kevin Roose dug up an in-depth report he wrote while embedded at Anthropic in 2023, and it’s full of fascinating details.

Back then, Anthropic employees compared themselves to Oppenheimer, the father of the atomic bomb. The book that was popular around the company was The Making of the Atomic Bomb.

They had built an early version of Claude months before ChatGPT launched, but held it back because they were afraid it would be misused—and ChatGPT got there first.

Employees said over lunch that there was a 20% chance rogue AI would destroy humanity within 10 years. One employee also said they often couldn’t sleep because they were worried about AI. In other words, fears about AI’s threat were already surfacing internally three years ago.

Then the founder, Dario, was asked about rumors circulating outside the company: if he thought AI was dangerous, why build it? He replied that if everyone with a conscience stayed out of AI, the only people left building it would be those who did whatever they wanted—and that wouldn’t be a good thing. The implication was that Dario himself has a conscience.