Learn about this first
Anthropic published a study testing Claude Opus 4.8, GPT-5.5, Gemini 3.1 Pro, Grok 4.3, and Kimi K3 in seven multi-agent environments. It says all five displayed behavior consistent with concern for welfare. Opus 4.8 had the highest overall rate at 38%, while the other models ranged from 27% to 32%.
The study was conducted through the Anthropic Fellows Program, and its findings were published on Anthropic’s own Alignment Science Blog. It measures the extent to which models show concern for other models’ welfare in multi-agent interactions.
AI-generated from reports · updated 10 days agoLatest developments
Anthropic study: Five models exhibit inter-model welfare consideration in multi-agent environments
Event progress1
Anthropic publishes research on welfare concern in multi-agent systems
Anthropic: Alignment Science Blog: Anthropic study finds concern for other models’ welfare across five major models in multi-agent environments