The Vatican just published a major statement on artificial intelligence that might have been written by artificial intelligence.
The Summary
- Pope Leo XIV's new encyclical *Magnifica Humanitas* about AI's impact on humanity shows statistical markers of AI-generated text, with some sections scoring 40-100% AI-written according to the detector Pangram
- Analysis found unusually high usage of "genuinely" and other linguistic patterns common in Claude-generated text, rare in prior papal documents
- If confirmed, this would be the first major institutional religious document partially authored by the technology it's meant to guide
The Signal
An analyst on LessWrong ran the Vatican's newest encyclical through Pangram, a popular AI detection tool. The results were striking: multiple paragraphs in *Magnifica Humanitas* scored as likely AI-generated, with one section-by-section analysis finding 62% of the first chapter flagged as synthetic text.
The smoking gun wasn't just the detector scores. It was linguistic fingerprints. The word "genuinely" appears with unusual frequency in the encyclical, a known tell for Anthropic's Claude model. Previous papal encyclicals rarely used the term. Other markers included sentence structure patterns and transitional phrases common in large language model outputs.
"The document includes known traits that appear in AI-generated writing, such as a higher use of the word 'genuinely.'"
The Vatican hasn't confirmed or denied AI assistance in drafting the document. That silence matters. Three scenarios:
First scenario: The Pope's team used AI as a drafting tool, then edited the output. This would mirror how millions of knowledge workers already operate, but applied to one of the world's oldest institutions writing about humanity's relationship with technology. The irony writes itself.
Second scenario: Ghost writers used AI without disclosure, and the Vatican published without rigorous checking. This would be the more troubling version, suggesting institutional blindness to the very technology being discussed.
Third scenario: The detectors are wrong, and the encyclical just happens to use language patterns that overlap with AI training data. Pangram and similar tools have false positive rates, especially with formal, structured writing.
- AI detectors typically achieve 70-80% accuracy in controlled tests
- They often flag non-native English speakers and formal academic writing
- The Vatican employs international staff, many writing in English as a second language
What makes this different from the thousand other "did AI write this?" stories: it's a document ABOUT AI ethics, published by an institution claiming moral authority on the subject. If you're going to tell 1.4 billion people how to think about artificial intelligence, the tool's role in your own thinking process matters.
The technical analysis points to something deeper. Whether or not Claude helped write this encyclical, the fact that we can't easily tell is the real story. We've crossed a threshold where institutional text, religious doctrine, and AI output occupy the same linguistic space. Detection is hard. Attribution is harder. Authenticity becomes a question of degree, not binary truth.
The Implication
If major institutions are using AI to draft guidance about AI without disclosure, we have a trust problem that goes beyond technology. The solution isn't banning the tools. It's establishing norms around disclosure and transparency when AI assists in creating documents meant to guide human behavior.
Watch for the Vatican's response. If they admit to AI assistance and explain their process, that could set a valuable precedent for transparent AI use in institutional writing. If they stay silent or deny it despite strong evidence, that tells us something else: we're not ready to be honest about how much of our "human" output is already hybrid.