Overview

Anthropic says it has spoken with scholars, clergy, philosophers and ethicists from more than 15 religious and cross-cultural groups. An Associated Press report published Sunday sharpened the question around one phrase: if a model is meant to reflect a “weighted average” of humanity’s moral views, how does it represent traditions carried through ritual, relationships and everyday practice?

A count shows reach, not influence

Anthropic said in a May post that it had met with people from more than 15 religious and cross-cultural groups and hoped to broaden those conversations. The AP report does not enumerate the traditions involved.

Philosopher Meghan Sullivan described a two-part meeting. The company first explained its models in plain language, then listened to feedback from faith leaders. Topics ranged from whether humans have souls to whether third graders should use a chatbot for literacy. Moral questions quickly touch product choices about capability, age and education.

The outreach is documented. Its effect on training, system rules and evaluation is not described in the report. A meeting is only one step between an invitation to speak and a change in model behavior.

Traditions do not arrive as rulebooks

Webb Keane, an anthropologist at the University of Michigan, argues that religion is entangled with culture, politics, doctrine, ritual and relationships. Extracting a comparable set of principles can make some traditions easier to represent while distorting others, especially faiths that do not center themselves on ethical rules.

Paolo Carozza, a legal scholar and member of Meta’s Oversight Board, worries that accumulated traditions could lose their defining features, with the result pulled toward a middle ground or even Western religions. Interviewees described this as a risk; the AP report did not present it as a measured outcome in a deployed model.

Keep the selection path visible

The number of groups consulted cannot establish representation by itself. Useful disclosure would explain who selects participants, which traditions or internal schools are missing, how disagreements are handled, and which model responses or evaluations reveal the effect of those choices.

A chatbot advising children on literacy and one answering whether humans have souls face different evidence and consequences. Folding both into “human morality” can hide value judgments behind a fluent answer.

The AP account confirms that Anthropic conducted broad consultations. It leaves open which models received input, how that input changes outputs, and how representation is assessed. Claims about a model’s moral grounding need observable examples of advice carried into behavior, along with evaluations that test the result.

Sources consulted

Associated Press, “To make AI morally good, tech companies are turning to religion. But some say it isn’t that simple,” published 2026-10-25 10:57:33 UTC. https://apnews.com/article/anthropic-moral-artificial-intelligence-religion-2bb91e3334fd6d732edf22288cabf4ec