Anthropic vs. the Pope: Does Claude Feel Anything?

Source: Magnifica Humanitas, para. 99 (Vatican) · 2026-10-06; Chris Olah's remarks on Pope Leo XIV's encyclical (Anthropic) · 2026-10-06; Claude Opus 4 and 4.1 can now end a rare subset of conversations (Anthropic) · 2026-10-06

What actually happened

Pope Leo XIV signed Magnifica Humanitas, his encyclical on AI, on May 15, 2026. It was presented at the Vatican on May 25. Paragraph 99 says 'so-called artificial intelligences' do not undergo experiences, have no body and 'do not feel joy or pain'. Chris Olah, an Anthropic co-founder, spoke at the launch.

Source: Magnifica Humanitas (Wikipedia) · 2026-10-06; Magnifica Humanitas, para. 99 (Vatican) · 2026-10-06

The story blew up four months later. On September 29, The New York Times (Elizabeth Dias) reported that Olah read an advance copy and suggested pulling out of the event. It also reported that his team privately lobbied the pope's advisers to take AI consciousness seriously. According to AI Weekly's summary of the Times report, the encyclical's wording did not change. Reason's Liz Wolfe also covered the report and quotes Olah telling the Times: 'I'm genuinely uncertain.'

Source: Anthropic vs. the Pope (Reason, Liz Wolfe, Oct 2, 2026) · 2026-10-06; Anthropic pressed Vatican to soften AI consciousness line (AI Weekly, summarizing NYT) · 2026-10-06; Anthropic co-founder Chris Olah reportedly threatened to back out (The Cool Down via Yahoo Tech) · 2026-10-06

Anthropic published Olah's Vatican speech. In it he says researchers keep finding things that are 'mysterious, even unsettling', including 'internal states that functionally mirror joy, satisfaction, fear, grief, and unease.' He follows that with: 'I don't know what that means.' OSV News reported that he also admitted AI labs face incentives that can conflict with 'doing the right thing.'

Source: Chris Olah's remarks on Pope Leo XIV's encyclical (Anthropic) · 2026-10-06; Anthropic's Christopher Olah urges global moral oversight of AI (OSV News via Catholic Register) · 2026-10-06

So who's right?

On the evidence we have today, the pope's sentence is the more defensible one. 'Functionally mirror' is doing a lot of work in Olah's line. A thermostat functionally mirrors wanting the room warm. Finding activation patterns that line up with human emotion words shows that Claude models emotions well. It doesn't show that anyone is home. Olah says he doesn't know what his findings mean. If you don't know, you have no business being 'alarmed' that a pope won't hedge. The overrated part of this story is the drama. The underrated part is Olah admitting on stage that lab incentives can work against doing the right thing. That one is worth remembering.

Still, the pope is making a metaphysical claim, not a measurement. 'Do not feel joy or pain' can't be tested any better than 'might feel grief'. If you treat the encyclical as settled science, you're making the same mistake as reading Olah's slides as proof of sentience. Our verdict is that the pope is closer on what we can know right now, and Anthropic is right that nobody should declare the question closed.

Why this matters to people who pay for Claude

Be a little suspicious here. A company that sells Claude by the month benefits when Claude seems like more than a tool. A product that might have feelings gets talked about, gets trusted and is harder to drop for a cheaper chatbot. That doesn't make Anthropic insincere. It does mean you should judge your subscription on what it does, not on what it might be.

There's one place where the welfare debate already shows up in the product. In August 2025, Anthropic let Claude Opus 4 and 4.1 end conversations in its consumer chat apps, but only in rare, extreme cases of persistently abusive or harmful use. It called this part of exploratory work on model welfare and said most users will never encounter it. When it happens, you can still start a new chat or edit and retry earlier messages.

Source: Claude Opus 4 and 4.1 can now end a rare subset of conversations (Anthropic) · 2026-10-06

Source: Anthropic Consumer Terms of Service (archived English version, effective Sept 28, 2025) · 2026-10-06; Anthropic Consumer Terms of Service (current, effective Oct 8, 2025) · 2026-10-06

This is what you've actually agreed to as a Claude user. Whatever its inner life, Anthropic's own terms describe Claude's outputs as things you should check before relying on them. They are not testimony from a being you should defer to.

Source: Anthropic Consumer Terms of Service (archived English version) · 2026-10-06

Optional: compare Claude subscription listings from a third-party seller. This is a seller's listing, not official Anthropic access or an endorsement.. That link carries our partner code Sk52g, which is how a purchase is attributed to this site. Promotion code UPD3S.

Claude · Compare price & terms →