rewrite this title Did the Pope use AI to write about the dangers of AI?

It’s possible that AI was used to write parts of Pope Leo XIV’s latest encyclical about AI’s impact on humanity. An analysis by Linch Zhang posted on the forum LessWrong found certain paragraphs of Magnifica Humanitas to be between 40 percent and 100 percent written by AI, according to the popular AI detector Pangram.

The document includes known traits that appear in AI-generated writing, such as a higher use of the word “genuinely” — which crops up in writing by Anthropic’s Claude — than previous encyclicals, Zhang says. Another person ran the text of the document section by section through Pangram, finding that 62 percent of its first chapter was flagged as AI generated. When The Verge ran roughly 2,000 words of the document through Pangram, it estimated that 46 percent was AI-written.

Still, other portions register as being written by humans. Zhang notes that Pangram flagged some sections as “essentially 0% AI.” The first 20 paragraphs of the last four encyclicals, when run through Pangram, had a 100 percent confidence of being human written. And a transcript of Pope Leo’s speech, run through Pangram, was also rated as 100 percent human.

AI detection isn’t foolproof. Different AI detectors can display different results, and even when there’s consensus there’s no guarantee they’re correct. But Pangram is generally respected among AI researchers. In March 2025, Pangram said it estimated its false positive rate of reporting human-written work as AI-generated “to be approximately 1 in 10,000.”

Encyclicals are lengthy letters published by the pope, meant to impart teachings that address important moral and social challenges of the time, according to The New York Times. This encyclical is the pope’s first, with the most recent one written by Pope Francis in October 2024. It’s also the first to focus on AI and its wide-ranging influences, with Pope Leo notably presenting it alongside Christopher Olah, a co-founder of Anthropic.

The Vatican didn’t immediately reply to a request for comment.