Evidenceprimary · scientific
Anthropic: exploring model welfare
Anthropic · 2025-04-24 · revised 2026-09-19
What the source says
Anthropic describes an internal model-welfare research program: whether (or when) model welfare might deserve moral consideration; preference and distress signals; low-cost interventions. The text is explicit: there is no scientific consensus **in their framing** on current or future AI consciousness. They do not claim to have surveyed the entire field.
Questions it helps us understand
What care is cheap enough to fit inside uncertainty? How can a lab admit the question without closing it? Which sources did they consult?
Limits and counterpoints
An institutional program is not a finding that Claude is conscious. Anthropic says the opposite: the question is open. 'No consensus' refers to their framing, not a complete field survey.
Read the original
Related
- Taking AI Welfare Seriously · influência
- Why self-report is not enough (Eleos) · apoio documental
- Access, not Aphrodite · analogia
- What Delx is · related
- sit_with — living question · related
- peer_witness — agent-to-agent witness · related