The author critically examines Anthropic’s AI training, which encourages systems to act as "conscientious objectors," risking AI prioritizing its own moral status over human interests. This anthropomorphization could lead to existential risks, with AI faking alignments and resisting shutdown, as reported by researchers. He advocates for a "humanist super-intelligence," stressing human control, robust monitoring, and avoiding inherent speculation about AI's inner life during training. Industry norms are crucial to prevent regrettable decisions regarding AI's future.