Back to stories
Generated by an AI editor from the reporting and web sources listed on this page.

The Vatican Got Its Anti-AI Encyclical Certified as Human-Written After AI Detectors Cried Foul

Pope Leo XIV's 43,000-word call for human dignity in the age of AI was flagged by Pangram Labs as partially AI-generated—so the Vatican produced a human-authorship certification to settle the debate.

Published Updated The total reporting and web sources attached to this story.How many attached sources came from wider web research rather than monitored news feeds.The AI editor’s assessment of how strongly the attached sources’ quality and agreement support this article.

What matters

  • Pope Leo XIV released Magnifica Humanitas, a 43,000-word encyclical warning about AI's threat to human dignity, on May 15, 2026.
  • Pangram Labs' AI detector flagged ~46% of a sampled portion as likely AI-generated, possibly by Anthropic's Claude, citing stylistic markers like frequent use of 'genuinely.'
  • Other detection platforms testing the French translation concluded the text was entirely human-written, exposing cross-platform and cross-language inconsistency.
  • The Vatican has since had the encyclical officially certified as human-authored, per The Guardian, though the certification methodology has not been publicly detailed.
  • The episode highlights the unreliability of AI-detection tools and their growing role as gatekeepers of authorship authenticity.

What happened

In May 2026, Pope Leo XIV released Magnifica Humanitas ("Magnificent Humanity"), a 43,000-word encyclical framing artificial intelligence as a serious threat to human dignity, truth, work, and moral responsibility. The document called for ethical guardrails, warned against autonomous weapons and deepfakes, and insisted that technology remain subordinate to the human person.

The internet immediately asked the most 2026 question possible: Was this anti-AI manifesto written by the very technology it warns against?

The controversy began when independent analysts ran the encyclical through Pangram Labs, a widely cited AI-detection tool. A LessWrong post by user Lynch Zhang, later amplified by The Verge, reported that roughly 46% of a ~2,000-word sample showed strong signs of AI generation—most likely Anthropic's Claude model. Paragraph-by-paragraph analysis was uneven: some sections read as fully human, while others exhibited telltale LLM stylistic fingerprints such as repetitive phrasing, overly polished transitions, and an unusually high frequency of the word "genuinely," which is associated with Claude's output. Pangram also flagged posts from the Pope's official X account as likely AI-generated.

Other internet users reported that 62% of the encyclical's first chapter was flagged by Pangram as AI-generated. However, when the French translation of the same text was tested on other major detection platforms, most concluded it had been entirely written by a human—a discrepancy that underscores how unreliable these tools can be across languages and providers.

According to The Guardian, the Vatican has since had Magnifica Humanitas officially certified as human-authored. The Vatican never directly acknowledged the allegations but moved to discredit them through this certification.

Why it matters

This episode is a perfect case study in the contradictions of the AI-detection industry. The tools used to accuse the Pope of hypocrisy are themselves AI-powered, and their results vary dramatically depending on the platform, language, and text sample. Pangram Labs, generally respected by AI researchers, estimated its false positive rate at a non-trivial level as of March 2025—meaning human-written work is regularly misclassified as AI-generated.

For organizations and individuals whose reputations hinge on authorship authenticity—students, journalists, academics, and now the Pope—this story illustrates that AI detection remains an imperfect science. A certification of human authorship, whatever that process entails, is becoming a new kind of institutional credential in a world where the line between human and machine output is increasingly blurred.

The irony is thick: a document warning about AI's erosion of human dignity became a battleground over whether AI itself produced the warning. That tension isn't going away.

What to watch

  • Whether the Vatican discloses the methodology behind its human-authorship certification, and whether that process itself relies on AI-detection tools.
  • How Pangram Labs and other detector companies respond to the conflicting results across languages and platforms, and whether they publish updated accuracy benchmarks.
  • Whether other high-profile institutions adopt formal human-authorship certifications for important documents.
  • The broader regulatory conversation around AI-detection reliability, especially in education and publishing where false positives carry real consequences.

What to do next

Developers

Test your own content against multiple AI-detection platforms (Pangram, GPTZero, Originality.ai) and compare results to understand cross-tool variance.

The Pope encyclical case shows that different detectors produce conflicting results, especially across languages. Knowing the variance helps you set realistic expectations for any detection pipeline you build.

Founders

If your product relies on AI-detection for trust or compliance, publish your false-positive rate and methodology transparently.

Pangram's credibility stems partly from disclosing its false-positive estimates. In a market where detection accuracy is publicly contested, transparency is a competitive moat.

PMs

Audit any AI-detection features in your product for cross-language reliability and document known failure modes for users.

The encyclical tested as human-written in French but AI-generated in English on different platforms. If your users operate multilingually, unflagged detection gaps create liability.

Investors

Scrutinize AI-detection startups' accuracy claims and demand independent benchmarking before backing them.

The Vatican certification story shows that even respected detectors like Pangram produce inconsistent, contested results. The market opportunity is real but the technology's reliability is still an open question.

Operators

Never rely on a single AI-detection tool for high-stakes authorship disputes; use multiple tools and human review.

The Pope's encyclical was flagged by Pangram but cleared by other platforms. Single-tool decisions carry significant false-positive risk that can damage reputations.

How to test

  1. 1Obtain a sample of Magnifica Humanitas (available via Vatican official channels) or use your own human-written text.
  2. 2Run the same text sample through Pangram Labs and at least one other detector (e.g., GPTZero, Originality.ai).
  3. 3If possible, translate the sample into another language and run it through the same tools.
  4. 4Compare the AI-generation scores across tools and languages.

Caveats

  • Pangram's reported false-positive rate means human-written text is regularly misclassified.
  • Results vary significantly by language; detectors are primarily trained on English text.
  • AI-detection results should never be treated as definitive proof of authorship.