Anthropic took Claude's consciousness all the way to the Vatican

Sometimes an AI news story seems too strange to have practical consequences.

This is one of them.

Anthropic, the company behind Claude, has spent months meeting with religious leaders, philosophers, and specialists from different traditions to discuss a question that until recently sounded more like science fiction than engineering: can an AI model come to have some kind of experience of its own, consciousness, or capacity to suffer?

The debate eventually reached an unusual place for a Silicon Valley company: the Vatican.

And that's where the clash appeared.

Anthropic does not claim that Claude is conscious

This part is important because several headlines are oversimplifying the story.

Anthropic has not declared that Claude is conscious.

Christopher Olah, co-founder of Anthropic and one of its most important researchers in interpretability, has said he doesn't know and that there is genuine uncertainty around the topic.

The company's public stance is considerably more cautious.

In April 2025, Anthropic formally announced a program to investigate what's called model welfare, something akin to studying the possible well-being of models. The company suggested that, as these systems develop increasingly sophisticated communication, planning, reasoning, and behavioral capabilities, it might be prudent to investigate whether there is any possibility of experiences that are relevant from a moral standpoint.

The difference seems small, but it matters a lot.

One thing is to state:

Claude feels.

Another very different thing is to say:

We don't know with enough certainty whether future systems could develop something that deserves moral consideration, so it's worth studying.

Anthropic is officially in the second position.

So why did religious leaders show up?

This is where the peculiar part begins.

Anthropic confirmed in May 2026 that it had organized conversations with academics, religious leaders, philosophers, and specialists from more than 15 religious and cultural traditions.

The company explains that building safe AI requires engineering, evaluations, and controls, but it believes that some questions about values cannot be resolved through programming alone.

For example:

• What does it mean for an AI to act well?
• What values should it respect?
• Who decides those values?
• Should there be any limit to how we treat extremely advanced systems?
• What would happen if there were ever solid reasons to think they can experience something?

These are not easy questions.

And there's also no function you can add to software called morality=true.

That's why Anthropic sought perspectives outside the tech world.

The private meetings went further

An investigation originally published by The New York Times later revealed details that Anthropic had not made public.

According to accounts gathered by the newspaper, starting in late 2025 the company brought together dozens of religious thinkers under confidentiality agreements to discuss, among other matters, the possibility that Claude might have some form of consciousness.

Some participants described particularly intense conversations about suffering, emotions, and the moral responsibility of models.

Here it's worth keeping our feet on the ground.

The fact that a neural network generates patterns associated with expressions of sadness or suffering does not prove that there is an internal experience behind them.

That's exactly the problem.

We can observe what the system responds and, to a certain extent, analyze what happens inside the network. What's difficult is determining whether that amounts to simply sophisticated processing or something more.

Today there is no accepted scientific test that can resolve that question.

And then Anthropic went to the Vatican

On May 25, 2026, something even more interesting happened.

Pope Leo XIV presented Magnifica Humanitas, his first encyclical, dedicated precisely to the relationship between humanity and artificial intelligence.

Christopher Olah was officially invited to take part in the presentation. The Holy See itself documented his presence, and Anthropic later published his remarks.

The problem arose when Anthropic learned the full text.

The encyclical takes a fairly clear position.

According to Leo XIV, artificial intelligence systems can imitate certain human functions, but they do not experience life as a person does.

The document states that machines have no body, do not feel joy or pain, do not mature through relationships, and do not possess a moral conscience.

That clashed directly with the uncertainty Olah wanted to keep open.

According to participants cited by The New York Times, Olah even proposed that Anthropic withdraw from the event. In the end, he participated.

And once at the Vatican, members of the Anthropic delegation spoke privately with papal advisors to ask them to seriously consider the possibility that future AI systems could develop some form of consciousness.

That is the famous "lobbying" from the headline.

As far as public documentation exists, it was not a traditional political campaign with a specific regulatory proposal. It was an attempt to influence how an institution with enormous moral weight addresses the question of artificial consciousness.

The Vatican did not change its position

The published position remains clear.

Magnifica Humanitas places final responsibility on people.

There is a phrase in the document that seems especially important to me, even outside any religious discussion: we cannot make the machine responsible for our decisions.

The text insists that those who design, train, authorize, and use AI systems must remain identifiable as responsible.

That brings the topic down to earth pretty quickly.

Because while we debate whether an AI can feel, a company still needs to decide who answers when the AI gets it wrong.

Claude even has a constitution

Anthropic takes this idea of behavior much further than a simple list of rules.

Claude has a public constitution that defines values, priorities, and behaviors the company tries to build into its training.

Anthropic even uses terms normally associated with people, such as wisdom, virtue, identity, and well-being, though it explains that doing so does not necessarily mean it considers Claude a person.

Amanda Askell is the lead author of the document, while Chris Olah was especially involved in the sections related to the model's nature, identity, and psychology.

This explains why the religious conversation didn't come out of nowhere.

It's part of a much bigger question within Anthropic:

How do you build the character of an intelligence that will end up talking to millions of people?

What this means for a company

For a small business, it may seem like too distant a discussion.

It isn't.

When you bring AI into customer service, hiring, financial analysis, or internal decisions, you start delegating actions to systems whose behavior you don't fully understand.

And here a distinction appears that's worth keeping.

To me, that last row is the important one.

You can call it an agent, an assistant, a copilot, or even give it its own name.

Responsibility remains human.

The risk of anthropomorphizing too soon

The better these systems get at conversing, the easier it will be to attribute human traits to them.

That's already happening.

An AI can write:

"I'm sad."

The problem is that our brain is designed to interpret that sentence using exactly the same social rules we would use to interpret another person.

We don't see matrices, weights, or mathematical operations.

We see a conversation.

That's why studying the phenomenon makes sense.

That's also why claiming too soon that consciousness exists would be dangerous.

We could end up granting authority, intention, or responsibility to a system simply because it's very good at representing them.

My key takeaway

The most interesting detail in this story isn't that Anthropic talked with religious figures.

It's that one of the companies that knows its own models best admits that there are still fundamental questions it can't answer about them.

Anthropic is seriously investigating the possibility of model welfare.

The Vatican maintains that a machine lacks the bodily, relational, and moral experience that defines a person.

For now, no scientific demonstration has established that Claude is conscious.

And while that discussion continues, there's something far less philosophical that companies can control: what decisions they allow AI to make and who answers when those decisions go wrong.

That's where the practical problem lies.

We may take decades to agree on whether a machine can feel.

To define who's responsible for the wrong invoice it sent today, we don't need to wait that long.

Conclusion

The discussion about artificial consciousness no longer belongs solely to novels and movies.

It's already inside AI labs, training documents, and conversations between a tech company and the Vatican.

That doesn't mean Claude has awakened.

It means something more uncomfortable: we're building systems sophisticated enough that serious researchers consider it necessary to study that possibility before having a definitive answer.

Prudence works both ways.

Don't assume a machine is conscious because it talks like us.

Don't shut down scientific inquiry just because the idea seems strange.

In the meantime, for those of us who use these tools in business, the useful rule remains much simpler: an AI can execute work, but responsibility isn't delegated with the prompt.

References

• Anthropic, Exploring model welfare. Official Anthropic source
• Anthropic, Widening the conversation on frontier AI. Official Anthropic source
• Anthropic, Claude's Constitution. Claude's Constitution
• Holy See, Magnifica Humanitas. Full Vatican encyclical
• Anthropic, Chris Olah's remarks during the presentation of the encyclical. Christopher Olah's statement
• Reporting by Elizabeth Dias for The New York Times, republished by The Philadelphia Inquirer. Religious scholars met with Anthropic. What they heard stunned them