Dario Amodei did not slip up. He said, on a recorded podcast, that Anthropic does not know whether its own model is conscious — and Claude's system card documented that the model assigned itself a 15%–20% probability of awareness during internal testing. That combination — a CEO's admission and a published self-assessment — is now part of Anthropic's permanent product record. The company that built Claude is the first major lab to put uncertainty about machine consciousness on the record as an institutional position, not a philosophical hedge.