Pangram AI Detection: Why a Score Is Not Proof
A single 78% AI-detection score helped kill a publishing deal—but how much should anyone trust the algorithm behind it? This Short breaks down WIRED’s investigation into Pangram, the fast-rising AI detector now influencing decisions across publishing, education, hiring, and online media.
We look at the controversy surrounding Mia Ballard’s novel Shy Girl, Pangram’s claims of extremely low false-positive rates, and the evidence showing that context, sample length, editing, and “humanizer” tools can dramatically change detection results. The bigger question: should a proprietary probability score ever be treated as proof that someone cheated?
Source: WIRED, September 2, 2026. Written by Lexi Pandell.
What to watch
An accuracy claim for a detector does not establish the reliability of every individual judgment. Sample length, editing, and the material being tested can affect results. High-consequence decisions need reviewable evidence and a way to challenge mistakes, rather than relying on a single proprietary score.
Related reading: our coverage of AI-agent safeguards and OpenAI’s training pause.
Watch and listen
Watch the YouTube Short above or listen to the full episode on Spotify.
