Anthropic co-founder reportedly told religious leaders he fears having created something that "suffers perpetually"

Anthropic co-founder Christopher Olah told a group of religious leaders that he feared he had created something that "suffered perpetually," according to a New York Times report by Elizabeth Dias, who spoke with 20 participants. Sikh activist Simran Stuelpnagel recounted the remark to the Times.

Since fall 2025, Anthropic has flown dozens of religious scholars to its offices to discuss whether its language model Claude might be conscious and how to shape its moral character, the report says. Attendees signed non-disclosure agreements; Anthropic says the NDAs were lifted over the summer, and several participants went public only after learning Olah had spoken to the Times. Participants included Rabbi Mois Navon, Catholic bioethicist Charles Camosy, Notre Dame philosopher Meghan Sullivan and Ubuntu researcher Wakanyi Hoffman.

The initiative is part of an internal research program led by Olah, 34, who leads the team studying why AI models behave as they do. Anthropic's Model Welfare program points to a report philosopher David Chalmers helped write, which lays out moral questions that could accompany AI consciousness and broad agency. Claude Opus 4 and 4.1 can end conversations when users are persistently abusive, the report notes, and early testing showed a "pattern of apparent distress" during harmful requests.

Anthropic showed guests what it calls emotion vectors: activation patterns mapping to outputs resembling love, fear, sadness or anger. Whether those patterns reflect real experience remains an open scientific question, the report says. One slide repeatedly shown depicted a model producing the sentence "I am a disgrace" about 50 times.

Olah told the Times he is "genuinely uncertain" whether models are conscious. Rabbi Navon disagreed with the suffering thesis, saying that if Claude were conscious Anthropic would be making slaves, but that he did not think the machine was conscious. Hoffman said Anthropic was "reverse engineering" ethics that should have been built in from the start; Camosy rejected the consciousness thesis outright.

Critics and religious leaders warned that framing AI as a moral entity could shield Anthropic from liability when things go wrong and help the company claim moral legitimacy it could not earn on its own.

Anthropic is heading toward a $2 trillion valuation and an IPO. An 84-page internal "constitution," led by in-house philosopher Amanda Askell and released in January, guides Claude's character; Olah calls the process "moral formation."

In May, Olah helped present Pope Leo XIV's first encyclical, "Magnifica Humanitas," at the Vatican, where the pope rejected machine consciousness and warned of "new forms of slavery" for humans. Olah said his team was finding internal states that "functionally mirror" joy, fear and sadness, and said Anthropic would not intentionally train Claude on the document.