News

An empirical framework for studying AI welfare has been published

Researchers from Eleos AI Research and NYU CMEP have published a paper proposing an empirical framework for measuring the question of whether AI systems can be subjects of welfare.

Compiled by Sergey KostenchukPublished 2026-08-12Updated 2026-09-18

2026-08-12 Research Anthropic

Expanded analysis for this story

Open the longform version with context, source trail, and what changed.

Read longform

Studying AI Welfare Empirically — Center for Mind, Ethics, and Policy Source

🤖 An empirical framework for studying AI welfare has been published

Eleos AI Research and NYU CMEP propose a measurable framework to answer the question — can AI systems be subjects of welfare.

🌍 The discussion moves from philosophy to empirics: probabilistic assessments, pluralism of evidence and independence from developers.

👤 The work provides a map — which properties to measure, in which systems and with what tools.

Source: https://nonhumanminds.org/studying-ai-welfare-empirically/

Sources

Story

Researchers from Eleos AI Research and NYU CMEP have published a paper proposing an empirical framework for measuring the question of whether AI systems can be subjects of welfare. The work moves the discussion on AI welfare from philosophy to the empirical plane — proposing specific measurement methods and three principles: probabilistic (welfare claims are probabilities, not binary yes/no), pluralistic (multiple types of evidence simultaneously) and the independence principle (assessments from organizations independent of AI developers).