X discussion: AP reports AI companies are turning to religion to teach chatbots right from wrong, with Anthropic citing talks with 15+ religious and cross-cultural groups

Background and the problem

Large models answer a huge number of value-laden questions every day, so what it means for an AI to be good is now a product question, not just a philosophy one. On Nov 1 the market-news account unusual_whales (@unusual_whales) relayed an Associated Press feature reporting that some AI companies are turning to religion as they work out how to teach chatbots right from wrong. The post says Anthropic has met with scholars, clergy, philosophers and ethicists from more than 15 religious and cross-cultural groups to help shape how Claude handles moral questions. It drew heavy discussion, and X listed it among its trending news topics. To be clear, this is a discussion sparked by a media report, not a new Anthropic announcement.

Architecture and how it works

Anthropic had already described this work officially. Its May 19 post, Widening the conversation on frontier AI, says the first round of dialogues was with wisdom traditions, and that the company hopes they inform the content of Claude's constitution, the values Claude is trained to embody and the range of behaviors it evaluates. It stresses this is not about aligning models with any one tradition's worldview, but about having Claude draw on religious, secular and political viewpoints with equal depth. The post also describes an experiment: inspired by the idea that a mentor or sponsor can act as an external conscience, the team gave Claude a tool it can call mid-task that returns a brief reminder of its own ethical commitments, and Claude tended to reach for it right before consequential actions.

Benchmarks and measured numbers

This is not a model launch, so there are no benchmarks. Anthropic says wiring the reminder tool into Claude's decision loop markedly lowered misaligned behavior on several internal alignment evaluations, but it published no figures and says it is still untangling whether the effect comes from the reminder itself or from pausing to reflect (vendor-reported). Verifiable buzz: as of Nov 1, 13:10 SGT, the post had about 879 likes, 189 replies, 66 quotes and 162K views.

Getting started

The discussion splits several ways: supporters say human traditions hold a lot of accumulated thinking on how character forms; skeptics worry about importing religious disagreements into AI or prefer open, revisable secular principles; and plenty of replies are jokes, such as one saying Claude's refusals will now sound genuinely disappointed in you. The practical idea for developers is the engineering pattern: before an agent takes a high-risk action, give it a callable values-reminder tool or a review step so it pauses and re-checks, then measure the effect with your own evals.