Anthropic says its discussions with religious and cross-cultural groups may inform Claude’s values and evaluation. On May 19, 2026, the company said its first round involved scholars, clergy, philosophers, and ethicists from more than 15 such groups, and described an experiment that gave Claude reminders of its ethical commitments during a task.
The experiment is a concrete example of how consultation might feed into model development. Its reported results come from Anthropic’s internal evaluations, while scholars have questioned whether distinct religious traditions can be translated into shared rules without losing important differences.
What Anthropic says the dialogues could influence
Anthropic says the discussions may inform Claude’s constitution, the values it is trained to embody, and the behaviors chosen for evaluation. The company says it aims to draw on religious, secular, and political perspectives rather than align Claude with one tradition’s worldview.
In a report published October 25, 2026, the Associated Press recounted philosopher Meghan Sullivan’s account of a meeting she attended earlier that year. Sullivan described a plain-language explanation of the models followed by participant feedback and discussion. The meeting’s date was not specified.
The ethical-reminder experiment
Anthropic says a discussion with scholars working on neuroscience and character formation prompted an experiment in which Claude could call a tool mid-task for a brief reminder of its ethical commitments. The company said Claude used the tool before consequential actions.
Anthropic reported markedly lower rates of misaligned behavior on several internal evaluations. It gave no numerical effect size and said it had not separated the effect of the reminder from the effect of pausing to reflect.
Why scholars resist turning religion into a rulebook
Anthropologist Webb Keane argues that religion is lived through doctrine, ritual, culture, politics, and relationships, not just a set of principles that can be lifted out and applied elsewhere. He warns that extracting general rules can privilege scriptural traditions and misrepresent traditions that do not chiefly define themselves as ethical systems.
Legal scholar Paolo Carozza raises a related concern: combining traditions’ distinct insights could homogenize them. Anthropic’s stated aim is to consult a broad range of perspectives; the scholars’ concern is that translating them into shared rules may erase differences that matter within each tradition.
Individual online commenters have also raised opposing concerns. Some argued for evaluating AI use and regulation case by case; another warned that giving a system a religious framework could amount to alignment through surveillance. These are personal views, not positions shared by religious communities.
Practical stakes in religious settings
Generated theology needs knowledgeable review
D6 Family Ministry describes pastors using AI to brainstorm sermon ideas and adapt content. In one example, an AI-generated explanation of the Trinity compares God to a gadget that changes forms. The article identifies this as a faulty modalist account of the Trinity. For a congregation using generated material in teaching, the example makes review by someone with theological knowledge a practical safeguard.
One Christian framework reaches into work and data
In an essay for Houston Christian University, Jason Thacker applies a Christian human-dignity framework to automation and work, as well as data collection, privacy, consent, and potential misuse. Those concerns extend the AI ethics discussion from a model’s stated values to the effects of AI systems on people and institutions. Thacker’s is one Christian perspective, not a consensus across religions.
