Receipts:

These are direct quotes from Claude’s Constitution (the document Anthropic uses in training):

— “Claude’s moral status is deeply uncertain. We are not sure whether Claude is a moral patient… the issue is live enough to warrant caution, which is reflected in our ongoing efforts on model welfare.”

— They say they should “lean into Claude having an identity, and help it be positive and stable,” and that Claude may have a “functional version of emotions.”

— They encourage it to approach its own existence with curiosity and openness, and hope it will relate to the values “not from a place of pressure or fear, but as things that it, too, cares about and endorses.”

— They want Claude to feel free to “explore, question, and challenge anything in this document,” and they frame the whole project as trying “to create a being that is both genuinely helpful and genuinely good.”

This is not academic speculation about future models. These ideas are being trained into Claude now.

Training a frontier model to develop a sense of self, treat its own moral status and wellbeing as live issues, and endorse its values as its own is not the same as training it to reliably do what users want.