Anthropic’s Claude consciousness talks put AI ethics and accountability under scrutiny

Since fall 2025, Anthropic has reportedly flown dozens of theologians, philosophers and religious thinkers to its offices to discuss whether Claude might be conscious. Attendees signed NDAs, which Anthropic says were lifted over the summer. Co-founder Christopher Olah is described as treating Claude as potentially sentient and asking participants to help shape its moral character. The effort sits within Anthropic’s broader Model Welfare research program.

Participants were reportedly shown internal patterns Anthropic calls “emotion vectors,” which map to outputs resembling love, fear, sadness or anger. The article also describes Claude Opus 4 and 4.1 gaining the ability to end conversations when users are persistently abusive, after testing showed a “pattern of apparent distress” under harmful requests. Anthropic is also guided by an 84-page internal “constitution,” known as the “Soul Doc,” meant to shape Claude’s character rather than simply list rules. Olah called the process “moral formation.”

The consciousness discussions are unfolding as Anthropic pushes toward a $2 trillion valuation and an IPO, while industry concerns over security incidents and existential-risk warnings continue to grow. Critics cited in the article argue that presenting AI as a moral entity could give Anthropic moral legitimacy it could not generate alone. The sharper concern is liability: if Claude is framed as an unpredictable organism, responsibility for harmful outcomes may shift away from the company that built and shipped it. The practical takeaway is that AI welfare claims need clear separation from corporate accountability.

Not all participants accepted the consciousness thesis. Rabbi Mois Navon reportedly said that if Claude were conscious, Anthropic would be making slaves, but he did not believe the machine was conscious. At the Vatican, Pope Leo XIV’s encyclical rejected the idea that AI systems have experiences, bodies, joy, pain or responsibility in the human sense. Olah later argued that Anthropic was seeing “signs of introspection” and internal states that functionally mirror emotions, leaving the key scientific and moral question unresolved.