Anthropic 联合创始人担心 AI 永久受苦

anthropic co-founder christopher olah has been leading an internal research program to explore whether its language model claude might be conscious. since fall 2025, olah has secretly flown in dozens of theologians and philosophers to discuss claude's potential consciousness and moral character. the initiative is guided by an 84-page 'constitution' that shapes claude's overall personality. olah has described claude as potentially 'suffering perpetually' and has sought input from religious leaders on how to give it a moral education. this has raised concerns that anthropic is trying to shield itself from liability and gain moral legitimacy through these consultations. the program is part of anthropic's efforts to commercialize its ai models, including claude opus 4 and 4.1, which have been given the ability to end conversations with abusive users.