Anthropic convenes global faith leaders to shape Claude’s moral compass — meetings held behind closed doors, reportedly raising transparency questions

Private summits with theologians

It has been reported that Anthropic held two-day, closed-door workshops at its San Francisco headquarters, inviting representatives from a wide range of faiths — including Confucianism, Daoism, Hinduism, Sikhism, Mormonism, Islam and major Christian denominations — to discuss the ethical foundations of its Claude family of AI models. The sessions, reported originally by Politico and summarized in Chinese press, reportedly followed a similar March convening with Catholic and Protestant clergy and were not open to the public; Anthropic is said to have covered participants’ travel and accommodation to allow undisturbed discussions.

What was discussed

Topics ranged from practical safety questions — how a model should respond to users expressing self-harm or suicidal intent — to more abstract debates such as whether a sophisticated model can ever possess spiritual or moral worth. It has been reported that some employees engaged deeply with clergy and ethicists, and that the company’s explainability team spent especially long sessions with Christian participants after internal work suggesting AI might display emotion had unsettled staff. A Silicon Valley Catholic priest involved in the talks, Brendan McQuill (布伦登·麦奎尔), has reportedly been collaborating with Anthropic on narrative-driven techniques to align Claude’s behavior with ethical norms.

Why this matters now

Anthropic’s outreach to religious authorities comes as U.S. AI firms face intense public and governmental scrutiny over safety, military ties and governance — it has been reported that Anthropic had a public dispute with the Pentagon earlier this year. Seeking moral legitimacy from faith institutions and the Vatican world of ethical engagement is an unusual but pragmatic route to shaping norms. For Western readers less familiar with China-facing coverage: Chinese outlets have flagged these moves because the convenings included religious traditions widely practiced in China, and because such private ethics-setting raises questions about corporate influence over what values are encoded into powerful AI systems.

The core tension

Can theology help teach a statistical model a useful, consistent “conscience”? Anthropic’s experiment is one answer to that question, but the secrecy of the sessions and the lack of public oversight also prompt concerns. Reportedly, the company wants moral frameworks that can adapt as models evolve — but who gets to decide those frameworks, and under what accountability, remains an open question.