Team surprised its writing model beat Pangram's AI detection without training for it

yawnxyz · x · 2026-11-16

Researcher andrewwhite01 discovered their writing model passed Pangram's AI-text detection — something the team never trained it to do, quipping "maybe we really did solve AI writing." Quoting julianharris notes he'd only seen one other system beat AI detection before, and it was closed source.

Besides being a fun capability anecdote, it casts doubt on AI detector reliability: a model not optimized against detection can pass naturally.

Original post →

More from Fun

Fun channel →