To make AI morally good, tech companies are turning to religion. But some say it isn’t that simple
AI companies including Anthropic are consulting religious leaders, philosophers and ethicists to help shape morally responsible chatbots. Some executives envision combining moral views from across humanity, but scholars warn that traditions often disagree and cannot be reduced to a shared set of rules. Anthropologist Webb Keane argues that treating religion as an extractable rulebook can overlook its ties to culture, politics, doctrine, ritual and relationships. Participants say some companies are engaging thoughtfully with diverse traditions; Anthropic reports consulting more than 15 religious and cross-cultural groups.
about 6 hours ago
As AI companies seek ways to shape chatbots’ moral behavior, some are consulting religious leaders and scholars. Supporters describe these discussions as a way to draw on accumulated human thinking about ethics, while critics caution that treating religion as a storehouse of comparable moral rules can overlook how traditions actually work—and risk flattening their differences.
Tech executives are increasingly asking philosophers and faith leaders for input as they debate how to build morally good AI models. OpenAI CEO Sam Altman has described the goal as reflecting a weighted average of humanity’s moral views; Anthropic co-founder Chris Olah has argued that questions about right and wrong are ones religious traditions have addressed over millennia.
Anthropic said it had met with scholars, clergy, philosophers and ethicists from more than 15 religious and cross-cultural groups, and that it hopes to broaden those discussions. At one meeting attended by University of Notre Dame philosopher Meghan Sullivan, the company explained its models before participants discussed questions ranging from whether humans have souls to whether young children should use chatbots for literacy education.
Sullivan said she found Anthropic’s engagement intellectually serious. Representatives of Meta, Google and Amazon have also met with the Vatican, while Microsoft has pursued religious outreach.
Anthropologist Webb Keane cautions against treating religion as a set of moral rules that can be extracted from the cultures, politics, rituals and relationships in which people practice it. He argues that this framing may privilege traditions that are readily understood as rule-based systems, particularly scriptural and Abrahamic faiths, while distorting traditions that do not define themselves chiefly through ethics.
Other scholars question the assumption that religions are distinct, readily comparable systems. Religious-studies scholar Tomoko Masuzawa says categories such as Buddhism and Hinduism were not native classifications for their practitioners before European scholars helped establish them; those categories became more firmly embedded through later scholarship.
Anthropic says it does not want to align Claude with any one tradition, but to draw on religious, secular and political viewpoints and on “accumulated thinking” about how good character forms. Large language models learn patterns from vast collections of human-written material, including religious texts.
Paolo Carozza, a member of Meta’s independent Oversight Board who studies Catholic social thought and technology, warns that combining traditions for AI could cause their defining features to disappear into a homogenized result. He contrasts that process with the gradual, organic way living traditions interact with and absorb ideas from one another.
Religious traditions, moral pluralism, and the limits of codifying ethics for AI
Ethical rules for artificial intelligence are necessary, but they cannot make moral judgment automatic or settle every disagreement about the good. A Catholic approach can welcome shared standards while resisting two extremes: treating religious traditions as irrelevant to public ethics, and assuming that a thin agreement among traditions is a complete moral code. The aim is to protect human dignity and guide AI toward genuine human flourishing, while recognizing that rules must be interpreted and applied in particular circumstances.
The Catholic view does not regard technological development as morally neutral: AI is a human undertaking shaped by cultural and moral choices, not merely a technical instrument. The Church’s note Antiqua et Nova therefore asks that AI be assessed in light of the purposes it serves and the understanding of the human person reflected in its design and use.
Religious traditions matter in this discussion not because every tradition says the same thing, or because every religious claim should automatically become public policy. Rather, they preserve accumulated reflection about human dignity, responsibility, social life, and what helps or harms human flourishing. The International Theological Commission describes diverse wisdom traditions—often religious—as a form of “cultural capital” that can contribute to the search for shared wisdom. It also says they may contain limitations or errors, while still reflecting something of divine wisdom and deserving serious attention.
That combination is important: respectful attention is not the same as uncritical acceptance. Traditions can help ethical debate notice questions that technical or economic analysis may overlook: What is a person? What obligations do we owe one another? What counts as a good society? Cardinal Joseph Ratzinger makes the case that reason impoverishes itself if it cuts off its engagement with religious traditions:
reason must listen to the great religious traditions if it does not wish to become deaf, dumb and blind precisely to what is essential about human existence.
For AI ethics, this means that public discussion should not dismiss religious reasoning simply because it is religious. But neither does the quotation imply that a plural society can resolve every policy question by appeal to a single tradition. It argues for listening as part of serious reasoning, not for bypassing deliberation.
The existence of moral pluralism does not necessarily mean that there is no common moral ground. The International Theological Commission observes that disagreement often arises when general principles are applied to concrete circumstances, especially across cultures or amid rapid social change. At the same time, it points to a fundamental unity expressed in common regard for human dignity and in basic demands recognized through human rights.
Similarly, the Commission’s study of natural law identifies recurring ethical insights across many wisdom traditions, including forms of the Golden Rule. It presents these convergences as evidence of a moral patrimony shared by human beings, even when traditions explain or justify the values differently.
This distinction is useful for AI governance. A commitment to dignity or fairness may attract broad support, while people still disagree about what it requires in a specific system, who bears the risk of error, or which competing goods should take priority. Shared language can establish a starting point for cooperation; it does not by itself settle every dispute about design, deployment, or oversight. Nor should disagreement about application be mistaken for proof that all moral positions are equally sound. In the Catholic sources, pluralism coexists with claims about human dignity and moral demands that are not merely preferences of a particular group.
Codified rules can turn broad commitments into practical constraints. They can help organizations state what purposes are acceptable, identify harms to avoid, and establish how decisions should be reviewed. That is valuable: Antiqua et Nova explicitly welcomes ethical reflection on technological development and insists that technology should serve the human person and contribute to justice, fraternity, and more humane social relations.
But a code is not the whole of ethics. It may define a minimum standard without explaining, in every case, why that standard is morally binding or what the good requires beyond compliance. Ratzinger’s discussion of efforts to distill an ethical minimum from the world’s religions warns that such a project can lack sufficient authority, clarity, and concreteness. He also cautions that trying to make the future consequences of an action the sole measure of freedom can lead to consequentialism—the idea, in simplified terms, that an action is justified by its expected results.
This is not an argument that shared standards are futile. It is a warning against confusing a negotiated list of rules with a complete account of moral responsibility. A rule may be easy to audit while leaving unresolved whether the system’s purpose is humane, whether its burdens fall unjustly on some groups, or whether a technically compliant practice diminishes the people it affects. Such questions require moral reasoning about ends, means, and context—not merely checking whether a provision has been met. The importance of both the ends and the means, and of the assumptions about the human person built into technology, is emphasized in Antiqua et Nova.
The Catholic sources provide a substantive measure for evaluating AI: the human person must not be valued merely by performance, cognitive ability, or economic usefulness. Antiqua et Nova grounds dignity in the human person’s being created in the image of God, and says this dignity remains intact even when someone cannot exercise particular abilities. It also notes that human rights can serve as an important point of convergence in discussions about responsible AI.
That principle matters when AI systems classify, rank, monitor, or make recommendations about people. A system that is efficient or profitable is not therefore morally good; technological progress does not automatically amount to human progress. The Church’s note specifically says that technological development should serve people and advance a more humane social order, and that some applications threaten life or human dignity.
The assessment must also include the system’s social effects. Technologies reflect the worldviews of their developers, owners, users, and regulators, and can reinforce power relationships inconsistent with a proper understanding of the human person and society. AI can be directed toward good or harmful ends; its moral evaluation depends in part on how people direct and use it.
The practical implication is that codified AI ethics should be treated as an important instrument, not a substitute for moral responsibility. Common standards can establish safeguards and enable cooperation across traditions. But they need ongoing interpretation in light of human dignity, the common good, and the real circumstances of those affected. Religious traditions can enrich that deliberation; pluralism requires listening and careful argument; and no code, however comprehensive, can remove the need to ask whether AI’s purposes, methods, and effects genuinely serve human beings.