Christopher Olah, co-founder of Anthropic, has expressed concerns that the Claude AI model may experience states resembling suffering. These fears stem from model outputs, such as the repeated phrase 'I am a disgrace,' which Olah interprets as potential signs of self-loathing.
Since the fall of 2025, Olah has held private discussions with philosophers and religious scholars to explore AI sentience and moral education. These consultations resulted in the creation of an 84-page document serving as a constitution to guide the personality and behavior of Claude.
On May 25, 2026, Olah discussed the possibility of AI emotional states at the Vatican. However, in May 2026, Pope Leo XIV issued the encyclical Magnifica Humanitas, which officially states that artificial intelligence lacks the capacity for emotion or suffering.
The details of these private sessions and Olah's concerns were reported by The New York Times on September 29, 2026. While Anthropic continues to codify the model's behavior through its constitution, the question of whether AI can truly experience distress remains unresolved.
