FULL STORY
Pangram AI Detector: From Bounty Challenge to Bias Controversy
A $500 bounty for pre-2020 human texts falsely flagged by Pangram went unclaimed, but the detector soon faced accusations of disproportionately flagging African writers. Users' tests suggest it may judge by publication year rather than content.
2026-11-13 ~ 2026-11-15 · 3 episodes · 10 posts
Episode 1 · Pangram Detects Zero False Positives on Pre-2020 Human Texts, $500 Bounty Unclaimed (2026-11-13, 4 posts)
A $500 bounty for any pre-2020 human text that Pangram flags as AI remains unclaimed: an economist tested 500 unpublished human articles and a user tested 599 more, with zero false positives—including African English samples, rebutting bias claims.
- $500 bounty offered for pre-2020 human text flagged as AI by Pangram — gleech · 2026-11-13
- Unclaimed $500 bounty: no pre-2020 human text has fooled Pangram's AI detector — FlorianGallwitz · 2026-11-13
- 500 pre-2020 human essays all pass Pangram, shoring up AI detector's bounty — TuhinChakr · 2026-11-14
- Tester ran 599 pre-2020 texts through Pangram, finding a zero false positive rate — TuhinChakr · 2026-11-14
Episode 2 · AI Detector Pangram Sparks Mislabeling and Bias Controversy (2026-11-15, 2 posts)
AI detector Pangram faces criticism for disproportionately flagging African scholars' writing and for issuing "100% AI" labels that cannot distinguish AI-generated from AI-edited text, which critics say is damaging writers' careers.
- Pangram AI detector can't tell generated from edited text, yet its "100%" label wrecks careers — ipeirotis · 2026-11-15
- AI Detector Dispute: African Writing Flags as AI, Sparking Bias Accusations — madeofAjala · 2026-11-15
Episode 3 · Users Question Pangram AI Detector for Judging by Year, Not Content (2026-11-15, 4 posts)
Users hypothesize that AI detector Pangram judges text by publication year rather than content, labeling pre-2020 text as human and post-2020 text as AI-generated. They also note that formal writing is more likely to be misclassified as AI.
- User hypothesizes Pangram AI detector defaults to writing year over content — madeofAjala · 2026-11-15
- Pangram AI detector may judge by writing year, not text, user argues — madeofAjala · 2026-11-15
- Neural nets latch onto year-content spurious links, thread argues — madeofAjala · 2026-11-15
- More formal writing gets flagged as AI more often, observer notes — madeofAjala · 2026-11-15