In an analysis published on 25 October 2026, the Associated Press wrote that tech companies are turning to religion and philosophy leaders to teach chatbots right from wrong. This is not a new product launch: at its centre are the meetings Anthropic announced in May 2026 and a presentation held at the Vatican in early 2026. The AP analysis brings those meetings together into a single picture and leaves a question open: how should millennia-old ethical traditions shape the behaviour of artificial intelligence models?
What Anthropic Did
In its article «Widening the Conversation on Frontier AI», published on 19 May 2026, Anthropic said that in recent months it had organized a dialogue with representatives of wisdom traditions. The first roundtables brought together scholars, clergy, philosophers and ethicists from more than 15 religious and intercultural groups. The company linked this work to a document called the Claude constitution: it defines the model's values and behaviour, and the dialogue is meant to help determine how those values should be shaped.
According to Anthropic, the goal is not to align models with the worldview of a single tradition. The company said Claude should draw on religious, secular and political perspectives with equal depth and rigour. At the meetings, the company first explained in plain language how its models work, and then religious leaders gave their feedback. As AP wrote, this order was followed at a meeting attended by Meghan Sullivan, professor of philosophy at the University of Notre Dame.
The questions covered a wide range: from whether people have souls to the practical question of whether a third-grader can learn literacy with the help of a chatbot. Anthropic said these conversations are still at an early stage and that further rounds are planned with lawyers, psychologists, writers and representatives of civic institutions.
The Vatican and Other Companies
In early 2026, Anthropic co-founder Chris Olah spoke at the Vatican at the presentation of Pope Leo XIV's encyclical on artificial intelligence. In the speech quoted by AP, he said:
"One lab cannot answer these questions... your traditions have carried them for millennia."
Anthropic is not alone. According to AP, representatives of Meta, Google and Amazon have also met the Vatican, and Microsoft is actively engaged in religious dialogue. OpenAI chief executive Sam Altman said last year that ChatGPT should be a weighted average expression of humanity's moral views. These statements converge on one point: chatbots are no longer just technical tools; they are becoming systems people turn to for moral advice, education and everyday decisions.
According to the AP analysis, companies began this dialogue not by chance, but because models now talk directly to millions of people. When a model answers a question, it does more than provide information — it also demonstrates a particular pattern of behaviour and values. That is why companies are turning to outside voices on how that behaviour should be shaped. At the same time, as AP notes, the dialogue itself has not yet settled the debate over which views should enter the model.
Criticism and Open Questions
The AP analysis also includes criticism. Webb Keane, an anthropologist at the University of Michigan, warned that treating religion as a set of rules could privilege some traditions and distort others. Paolo Carozza, a member of the Meta Oversight Board, said that blending different traditions could strip them of their distinctiveness and pull them toward an average value.
These two objections show one question from different sides. The first objection concerns the choice itself: if one tradition is adopted as a rule, other traditions may be left aside. The second concerns the result of blending: taking a little from every tradition may fail to preserve the full essence of any of them. The AP analysis cites no specific response from the companies to these objections.
The tension becomes clear when Anthropic's position is placed alongside the criticism. The company said it aims to draw on all perspectives with equal depth, without privileging one tradition. Critics argue that this very breadth carries a risk: the wider the selection, the more each tradition's distinctiveness may blur.
This material should therefore be read not as today's news, but as an analysis published today of a process unfolding this year. On 25 October 2026, AP announced neither a new product nor a new rule; it merely gathered into one analysis the meetings, statements and objections accumulated over 2026. Questions remain open: which voices companies will listen to, which views they will build into model behaviour, and who will be accountable for that choice.
