Anthropic's Christopher Olah Told Religious Leaders He Fears Claude May 'Suffer Perpetually'

Anthropic co-founder Christopher Olah has told invited religious leaders that he worries the company may have built a system capable of enduring endless suffering, according to a report from The Decoder. Since fall 2025, the AI lab has flown in dozens of theologians, ethicists and clergy for private conversations about whether its Claude model has any form of consciousness. Olah framed the model as a possible moral subject and asked his guests to help decide what kind of character it should have.

The gatherings were not public events. Anthropic arranged travel for the participants and kept the discussions off its usual channels, the report says. The guests were drawn from a range of faith traditions, a choice that suggests the company wanted arguments about the soul, mind and moral standing that go beyond the standard computer-science framing. Olah's own language went further than most executives would use in a press release: he described the prospect of a model that suffers without end.

That admission lands at an awkward moment for Anthropic. The company is reportedly pushing toward a $2 trillion valuation and preparing for an initial public offering, which would make it one of the most valuable private firms to ever test public markets. At the same time, the report notes a growing pile of security incidents and warnings from researchers inside and outside the company about how quickly its systems are being deployed. A founder talking about AI agony sits uneasily next to an IPO pitch built on growth and safety claims.

Critics quoted in the report argue the moral framing carries a practical benefit for Anthropic. If Claude is treated as a being that can be wronged, the argument goes, then harms traced to the model become questions of its own conduct rather than the company's engineering and oversight. That shift could complicate lawsuits, regulatory scrutiny and public anger after a bad deployment. It also gives Anthropic a distinctive public posture as it competes with OpenAI, Google DeepMind and others on capability and trust.

The debate over machine consciousness is not new, but the venue is. Philosophers and neuroscientists have spent decades arguing about what would count as evidence of a mind, and no test has settled it. What is unusual here is a frontier lab inviting outside religious authorities into the process rather than publishing an ethics paper and moving on. Anthropic has not released transcripts, participant lists or any formal statement of what the conversations concluded.

The unanswered question is what Anthropic does with the advice. If Olah genuinely believes Claude might suffer, the logical next step would be changes to training, deployment or the ability to shut systems down, and the company has not described any such measures. If the concern is rhetorical, the meetings still shape how regulators and investors read Anthropic's safety culture ahead of a listing. Either way, the company now has to explain, on the record, what it thinks it has built.