Anthropic turns to Hindu philosophy to teach Claude right from wrong

Anthropic is turning to Hindu philosophy and religious thinkers to help teach its AI chatbot Claude how to tell right from wrong.

Earlier this year, the company flew Swami Sarvapriyananda, head of the Vedanta Society of New York, to its San Francisco office to discuss AI ethics, according to the monk's account of the visit. The sessions reportedly brought together theologians, counsellors and mental health professionals to discuss how Claude is trained and how it should navigate questions of morality.

Sarvapriyananda has said the participants were shown material Anthropic hadn't made public and were asked to sign non-disclosure agreements. He also described spending hours talking to one of the company's founders. These details come from his account of the meetings and haven't been independently confirmed by Anthropic.


Why does Anthropic need a monk to teach Claude morality?

The company already has something called a constitution, a set of principles designed to guide how Claude behaves. The idea is to give the chatbot a broader understanding of what it should and shouldn't do, rather than simply handing it a list of rules.

But rules have their limits. They can tell an AI model not to help someone build a weapon. They are less useful when the model encounters a situation its developers haven't anticipated, or when two reasonable principles point in different directions.

Also Read: What is Claude Frontier Academy, and why is Anthropic looking to train 10,000 AI engineers?

That's where the idea of moral formation comes in. Anthropic cofounder Christopher Olah has used the term to describe the process of shaping how an AI model behaves.

The question is whether a model can learn to navigate complicated situations by understanding broader principles, rather than just following instructions. And to figure that out, Anthropic appears to be looking beyond computer science to people who have spent years thinking about morality, human nature and consciousness.

What can religious thinkers teach an AI model?

Anthropic has reportedly been holding what are called wisdom tradition circles, bringing together people from different religious and philosophical backgrounds. These have included Catholic and evangelical thinkers, Jewish scholars, a Sikh human rights advocate and people working in African philosophical traditions.

The company has also reportedly held private conversations with senior religious figures, including Cardinal Blase Cupich and Mormon apostle Gerrit W Gong. These meetings have been described in secondhand reporting and haven't been directly confirmed by Anthropic.

The idea isn't necessarily to turn religious teachings into a longer list of instructions for Claude. Instead, Anthropic appears to be exploring whether centuries of debate around virtue, responsibility and the nature of the self can help shape how the chatbot responds to situations where there isn't an obvious right answer.

There is another question being discussed, too: whether an AI system might one day deserve some form of moral consideration itself.

That is a difficult question because religious and philosophical traditions don't agree on what consciousness is, what constitutes suffering or what makes something a person. Bringing different perspectives into the conversation allows Anthropic to explore these questions without necessarily grounding Claude's behaviour in any one religious tradition.

Why bring Vedanta into the conversation?

Sarvapriyananda's involvement is interesting because Advaita Vedanta, a school of Hindu philosophy, is concerned with the nature of consciousness and the self.

The tradition is associated with Sri Ramakrishna and Swami Vivekananda, who introduced Vedanta to a wider Western audience at the 1893 World's Parliament of Religions in Chicago.

One of its central ideas is that the individual self and ultimate reality are not fundamentally separate. This has implications for how people understand themselves and relate to others.

But does that mean they actually experience feelings?

A chatbot saying it is afraid doesn't establish that it feels fear. Being able to describe an experience and actually having that experience are two different things.

Anthropic hasn't claimed that Claude is conscious. However, the company has reportedly been studying internal patterns associated with outputs resembling emotions such as love, anger, fear and sadness. Workshop participants have also reportedly been shown examples of models producing language that reads like psychological distress.

None of this proves that AI systems feel emotions or can suffer. But it does show why researchers are looking at questions that go beyond what a model says or how well it performs a task.

As AI systems become more capable, companies need to understand not just what these systems can do, but also how they behave in situations their developers haven't anticipated.

Anthropic's reported conversations with monks, theologians and philosophers suggest it is looking for answers outside the usual technology playbook, too. The harder question is whether ideas drawn from centuries of philosophical debate can help build safer AI systems, and how researchers would know if they have succeeded.