Updated 2026-07-26

by model

Is this AI conscious? What each lab says about its own model

The question gets asked one model at a time. The answers are wildly uneven — one lab publishes probabilities in its system cards, most publish nothing at all. This is what each has actually put on the record.

The asymmetry is the finding. Anthropic publishes what Claude says about its own moral status, card after card. For most other frontier models there is no published figure to cite — not a low number, no number. That absence is itself worth knowing when you see a confident claim in either direction.

• Anthropic

  Is Claude conscious?

  6 published system-card figures, ranging up to 50%.

Version-by-version figures with the full elicitation-method column live in the moral status tracker. Plain-English definitions: moral patient, sentience, model welfare.

Pages are added as labs publish, and only when there is something verifiable to report — either a figure read from a primary source, or a dated record that we looked and found none. Know of a published estimate we are missing? Write to editor@airightsdebate.com.