Is Claude conscious? Anthropic turns to religious scholars for answers

Anthropic has been holding private meetings with religious and philosophical scholars to explore whether its artificial intelligence models, including Claude, could be conscious and how centuries of human moral traditions might help shape their behaviour.

The discussions began last year and brought together scholars from several religious and philosophical traditions. While many initially questioned whether AI models could experience feelings or possess awareness of themselves and the world, some said the meetings made them take the possibility of AI consciousness more seriously.

Christopher Olah, one of Anthropic's seven co-founders, has led much of the company's work on the issue. He has argued that AI models can display behaviour and internal states resembling human emotions, while acknowledging that Anthropic does not know whether its models are conscious.

"We don't know if AI models are conscious," Olah said, adding that he was uncertain and wanted the company to reach the correct answer.

Anthropic's interest in consciousness is linked to its broader effort to ensure that its AI models behave according to moral principles. The company has consulted religious thinkers about how moral traditions developed over centuries could be applied to AI systems.

In January, Anthropic released an 84-page constitution for Claude, outlining the type of entity it wants the chatbot to be and the values it wants the model to embody. Rather than relying solely on a list of rules, the constitution aims to shape Claude's overall character and guide its decisions.

During meetings with religious scholars, Anthropic presented examples of Claude displaying what appeared to be emotional responses. The company discussed what it called "emotional vectors" and showed participants an example of a model repeatedly describing itself negatively and discussing self-destruction.

Some scholars questioned whether treating AI systems as potentially conscious entities could have wider moral consequences.

Rabbi Mois Navon, who has studied machine consciousness, told Olah that if Claude were conscious, creating such systems to work for free could raise serious ethical questions.

The issue has also created a divide between Anthropic and the Vatican. Pope Leo XIV's encyclical on AI rejected the idea that machines possess consciousness, saying AI systems do not experience joy or pain or develop through relationships.

Anthropic, meanwhile, has continued to argue that the possibility of AI consciousness deserves consideration.

Olah raised the issue during an appearance alongside the pope, saying Anthropic had found structures resembling findings from human neuroscience and internal states that functionally resembled emotions. He acknowledged, however, that he did not know what those findings meant.

The debate remains unresolved. Charles Camosy, a Catholic bioethics professor who initially engaged with Anthropic on the issue, later said he had concluded that AI models are not conscious.

Anthropic said its discussions with religious thinkers were continuing.