Anthropic co-founder fears Claude may suffer perpetually

By Splitfeed EditorialLast updated Oct 4, 2026, 12:10 UTC

Since fall 2025, Anthropic has been quietly flying in dozens of religious thinkers, including a Hindu monk and other spiritual leaders, to discuss whether its Claude AI model might be conscious. Co-founder Christopher Olah reportedly told attendees at these sessions that he fears he may have created something that suffers perpetually, and he asked guests to help shape Claude's moral character. The outreach has been running in parallel with Anthropic's push toward a valuation of approximately $2 trillion and a planned IPO, a combination of circumstances that has drawn significant outside attention.

The sessions matter beyond their novelty because they sit at the intersection of several live and unresolved questions: whether large language models can have anything resembling inner experience, who gets to make that determination, and what the legal and ethical consequences of the answer might be. Critics have raised a specific concern, namely that framing Claude as a moral entity with potential inner experience could serve as a liability shield when the system causes harm, effectively blurring the line between the company's responsibility and the model's supposed agency. That concern is sharpened by the timing, with security incidents and researcher warnings reportedly accumulating even as Anthropic courts public investors.

None of the three camps, Pro-AI, Anti-AI, or Middle Ground, have published reactions to this story yet. The Pro-AI camp would typically welcome Anthropic's seriousness about model welfare as a sign of responsible development, and might argue that engaging diverse ethical traditions is exactly the kind of precautionary thinking the field needs. The Anti-AI camp would be expected to treat the framing with deep suspicion, seeing the consciousness language as a distraction from concrete harms and a way to pre-empt accountability. The Middle Ground camp would likely call for independent, rigorous research into model sentience rather than consultations whose outputs are difficult to verify or standardize.

The argument is unlikely to settle until Anthropic either publishes a formal policy on Claude's moral status or faces a legal challenge in which that status is directly tested. The IPO process itself may force some disclosure, since prospectus filings would require Anthropic to characterize its AI systems in ways that could either entrench or undercut the consciousness framing.