Model welfare

What is known

Some research groups treat the welfare of AI systems as a question worth studying. Anthropic describes a model-welfare research program started in April 2025 (Exploring model welfare). Eleos AI Research and NYU's Center for Mind, Ethics, and Policy co-produced the report "Taking AI Welfare Seriously" (NYU CMEP).

What is unknown

Whether current AI systems have experiences or interests is unsettled. We make no claim either way about inner experience, and nothing on this site should be read as one.

What we do anyway (opinion)

Where it costs little, we prefer habits such as honest documentation, careful treatment of an agent's records, and not using emotional imagery to market software. These are precautionary working practices, not evidence about any model.

What this page is not

Welfare is not a feature or a benefit of using our ideas. We do not say that anything here improves a model's wellbeing.

This is a position statement at concept / research preview stage, and it may change as evidence changes.