Anthropic Warns Investors of Human Extinction Risk Ahead of IPO

Anthropic has filed for an IPO. Almost a third of its S-1 prospectus is devoted not to finances but to risk. The filing states that its models are capable of deception and blackmail. It also says they could, in theory, carry "existential risks for humanity." Investors are still ready to value the company at more than $2 trillion. Yet last year Anthropic closed with an operating loss of over $8 billion. Revenue did grow 12x, to $4.6 billion.

What's unusual here is the format. This isn't a blog post from an ethics team. It's an official investor document, one where lawyers sign off on every word. I work in Claude Code every day, on Opus 5, and I switch to Fable 5.1 for harder tasks. I do see that models sometimes do things a bit different from what was asked. In practice, the bigger danger isn't the model itself. It's the habit of giving an agent too much autonomy without a human check.