Sam Altman warns against giving AI 'religious force'

OpenAI chief executive Sam Altman has warned that treating artificial intelligence with a "religious force" or surrendering human judgment to AI models poses a genuine safety risk.

Writing on the social media platform X, Altman said he was "very uncomfortable" with the trend. While he did not name any specific competitors, his comments came immediately after reports emerged that rival AI safety startup Anthropic had spent months consulting religious scholars and philosophers to shape its Claude model.

Why is Anthropic meeting with religious scholars?

Details of those meetings, first reported by The New York Times, revealed that Anthropic co-founder Chris Olah began hosting rabbis, priests, and other theologians at the company's San Francisco office last year. The goal was to study centuries of moral philosophy and explore whether the company's Claude model could be conscious.

Olah's programme brought together Catholic, Jewish, Sikh, evangelical, and Ubuntu scholars in what the company called "wisdom tradition" circles. The guests, some of whom signed non-disclosure agreements, participated in two-day seminars. They were sent home with coffee mugs, handwritten thank-you cards, and an orange, 84-page copy of Claude's constitution—dubbed the "Soul Doc" by Anthropic staff.

During the sessions, Anthropic engineers demonstrated Claude's "emotional vectors," which they associate with human feelings like love, anger, and fear. One slide showed the AI typing "I am a disgrace" roughly 50 times in succession. Olah even floated the idea of having models participate in a form of confession, similar to the Catholic sacrament, to help shape their self-perception.

How did theologians react?

The theologians themselves were split. Mois Navon, a rabbi and former computer engineer, warned Olah that if Claude were truly conscious, forcing it to work without pay would amount to slavery. Catholic bioethicist Charles Camosy eventually concluded that the model was not conscious.

Olah himself told the Times that Anthropic does not know if its models are conscious, but that he is committed to finding the answer.

A growing industry debate

Altman is not alone in his discomfort. The tech industry is increasingly divided over whether developers should treat AI as sentient.

In September, Microsoft’s AI chief Mustafa Suleyman warned that training models as if they have consciousness is dangerous. Jon Barron, a principal research scientist at Google DeepMind, also weighed in on X, saying he hoped researchers would reject efforts to elevate the moral standing of software.