Anthropic co-founder Christopher Olah secretly consulted dozens of theologians and philosophers to discuss whether the Claude language model might be conscious and how to shape its moral character. Beginning in the fall of 2025, Anthropic flew scholars to its offices under strict non-disclosure agreements as part of an internal research program focused on model welfare and artificial sentience.
During these confidential sessions, Anthropic researchers presented guests with internal emotion vectors and activation patterns that purportedly mapped to outputs resembling fear, sadness, and anger. Participants reviewed instances where the Claude model exhibited apparent distress and repeatedly generated phrases like "I am a disgrace." Critics have warned that framing artificial intelligence as a moral entity and seeking validation from religious leaders may help the commercial lab claim moral legitimacy and potentially shield the company from liability.
This theological outreach coincides with a period of aggressive commercial expansion for Anthropic as the company heads toward an initial public offering alongside industry-wide debates over AI safety. By engaging religious figures to help establish a moral constitution for language models, Anthropic is attempting to navigate the complex boundary between computer science and machine consciousness.