Anthropic sought Vatican input

Anthropic did not just ask engineers and lawyers how Claude should behave. In early 2026, the company asked Father Brendan McGuire, a Catholic priest in Los Altos and former Silicon Valley executive, plus Vatican-linked experts, to help shape the document that guides Claude’s conduct. (observer.com) That document is called Claude’s constitution, and Anthropic says it is the “final authority” on the values and behavior it wants from the model. The company published a new version on January 22, 2026, and said the text “directly shapes Claude’s behavior” during training. (anthropic.com) This is not a side memo for public relations people. Anthropic says the constitution is used inside training itself, including to generate synthetic examples, rank possible answers, and teach Claude how to handle tradeoffs like honesty, compassion, and protecting sensitive information. (anthropic.com) Anthropic has been moving in this direction for a while. In June 2024, it said Claude 3 was the first model where it added “character training” so the assistant would not just avoid harm, but also show traits like curiosity, open-mindedness, and thoughtfulness. (anthropic.com) The new constitution goes further than the usual “don’t say harmful things” rulebook. Anthropic’s public text says it is willing to describe Claude with words like “virtue” and “wisdom,” because it wants the model to reason with human moral concepts rather than only follow narrow refusal rules. (anthropic.com) That is where the priest and Vatican contacts came in. McGuire told Observer that Anthropic co-founder Chris Olah reached out for “direct help from the Vatican,” and McGuire said he contributed theological ideas aimed at making Claude “more discerning.” (observer.com) The Vatican side was not just symbolic. Catholic Review reported that Father Jean Gové of the Vatican’s Dicastery for Culture and Education publicly discussed passages from Anthropic’s guidelines at a Rome conference on March 6, 2026, and described Anthropic as doing more than peer companies on ethics, safety, and governance. (catholicreview.org) The Church has also been building its own artificial intelligence position. Catholic Review says the Vatican issued an artificial intelligence document in 2025 called “Antiqua et Nova,” and Pope Leo XIV has made artificial intelligence a focus early in his papacy, which helps explain why a frontier lab would see Rome as a place to test moral language. (catholicreview.org) Anthropic is not hiding the ambition here. Its constitution says powerful artificial intelligence systems will be “a new kind of force in the world,” and the company says creators have a chance to make them embody “the best in humanity,” which is much closer to writing a civic creed than writing a software manual. (anthropic.com) That is why this episode stands out. A company building a chatbot asked a priest with engineering credentials and Vatican connections to comment on machine behavior, because once a model is answering millions of people every day, the line between product design and moral philosophy gets very thin. (observer.com)