Anthropic’s Quiet Push to Make the Vatican See AI as Conscious

The explicit disagreement between Anthropic and the Vatican reveals a fundamental tension over who defines the ethical boundaries of advanced AI. While the Vatican, through its encyclical "Magnifica Humanitas," drew a clear line against machine consciousness, Anthropic's internal discussions with scholars suggest a willingness to seriously consider the possibility of AI models like Claude being sentient. This divergence highlights that even partners with shared safety goals can hold deeply different views on AI's inherent nature.

Anthropic's outreach to religious scholars, philosophers, and ethicists, initially under NDAs, indicates a proactive effort to explore complex ethical questions beyond purely technical considerations. The company's focus on "interpretability work" and Olah's statements about finding "mysterious, even unsettling" internal states within models suggest an internal drive to understand AI's emergent properties, even if it challenges established theological positions.

The Vatican's insistence on an absolute boundary between human and machine, coupled with its warning against ceding moral authority solely to engineers, points to a broader concern about the unchecked power of AI developers. The encyclical's call for "disarmed" AI and rigorous ethical constraints, especially regarding military applications, underscores a desire to embed external values into AI governance, rather than allowing developers' internal moral visions to become the "invisible infrastructure" of these systems.

This episode illustrates that the debate over AI's moral status is not merely philosophical but has practical implications for governance and regulation. The clash between Anthropic's exploratory stance and the Vatican's definitive theological position shows that even when companies align with external bodies on high-level principles like safety, deep-seated disagreements about AI's fundamental nature can complicate collaborative efforts to establish ethical guardrails.