When AI lobbies the Vatican

Anthropic co-founder Christopher Olah and his colleagues tried to convince even the Pope's advisors to take seriously the possibility of artificial intelligence consciousness, after the preliminary text of Pope Leo XIV's first encyclical clearly rejected this idea. According to a report by Elizabeth Dias, a religion correspondent for The New York Times, the Vatican consultation was preceded by months of closed-door meetings in which Anthropic discussed the moral status of its own Claude models with religious scholars and philosophers.

The company began consulting representatives of various religious traditions in the fall of 2025, and later invited dozens of participants to its San Francisco offices. The discussions were covered by confidentiality agreements, which Anthropic says were lifted this summer. Dias spoke with twenty religious and philosophical thinkers about the meetings; the company said it had involved representatives of more than fifteen religious and cultural traditions in the dialogue. The initiative served both to examine how moral principles could be built into the models' operation and to debate whether the systems could have their own experience or consciousness. Several participants, however, felt that during the conversations particular emphasis was placed on Claude's possible suffering and its protection.

Rabbi Mois Navon, who previously worked as a computer engineer and wrote his dissertation on the ethics of machine consciousness, said that Anthropic's leaders treated the model as a conscious being. At the presentations they spoke of "emotional vectors" and showed an example in which a model repeated, without stopping, roughly fifty times: "I am ashamed." Sikh human rights activist Simran Jeet Singh Stuelpnagel recalled that Olah feared he had created a system that was suffering continuously.

From this, Navon drew a conclusion uncomfortable for Anthropic: if Claude were truly conscious, then putting it to work without pay would amount to slavery. The rabbi himself, however, does not consider the model conscious. Olah told the Times that he, too, does not know whether AI systems possess consciousness; what matters to him, he said, is finding the right answer.

The debate continued in May at the Vatican, where Olah was invited to the presentation of the encyclical Magnifica humanitas. According to one Vatican organizer, the document he received in advance unsettled him so much that he raised the possibility of Anthropic withdrawing from the event. In the end they did attend the May 25 presentation, but according to two participants, beforehand Olah and his team tried, in one-on-one conversations, to persuade the Pope's advisors to leave the question open.

The encyclical's position did not change: according to the document, artificial intelligences do not have experiences, have no body, feel neither joy nor pain, and possess no moral conscience. The text emphasizes the protection of human dignity and warns that, without jointly established ethical frameworks, the values of those who control AI could become the invisible foundation of the systems. Yet at the presentation Olah returned to the question of consciousness, citing research findings that he said point to introspection, as well as internal states that function similarly to joy, fear and other emotions. At the same time, he acknowledged that the business incentives of Anthropic and other leading AI labs could clash with ethical conduct, which is why he also called for external accountability.

The initiative drew criticism from several quarters. Mustafa Suleyman, head of Microsoft's AI division, warned that belief in machine consciousness could carry social risks; Charles Camosy, a Catholic bioethicist who took part in the meetings, later concluded that the models are not conscious; and Wakanyi Hoffman, a researcher of Ubuntu philosophy, objected that Anthropic was trying to retrofit ethics onto a technology that had already been created, when such questions should have had a place at the design stage.

Fallow Ground