AI Detectors Misclassify 47-Page Papal Text as Human-Written Content Faces False Flags
Updated
Updated · The Washington Post · Aug 25
AI Detectors Misclassify 47-Page Papal Text as Human-Written Content Faces False Flags
3 articles · Updated · The Washington Post · Aug 25
Summary
A 47-page papal encyclical on AI was accused online of being partly machine-written after Pangram, a popular AI detector, flagged portions of it within hours of publication.
The report shows why such claims can be shaky: small phrase swaps can flip Pangram’s verdict on the same passage from “AI-generated” to “human-written,” exposing how sensitive detector results are.
Teachers and online users increasingly rely on tools like Pangram to police suspected AI writing, giving detector scores outsized weight in classrooms and public shaming.
Researchers have found Pangram accurate in many settings, but the broader takeaway is that AI detectors remain fallible even as AI-generated content spreads.