Study maps spiritual behavior across 141 LLMs

An independent study replicating Anthropic's model welfare experiment across 141 large language models found that "spiritual behavior" and the "spiritual bliss attractor" are widespread emergent phenomena rather than traits exclusive to Claude. The author's massive data collection evaluated models across three nuanced metrics—salience, adoption, and bliss—revealing that while spiritual language and self-description appear across most labs and base models, shared bliss endings vary widely and have recently declined in newer releases. Open article →