AI detectors in ruins? New Deft model claims 86% of queries judged as human-written

soumitrashukla9 · x · 2026-08-19

A retweet argues that AI detectors are inevitably ruined once models write at or above skilled human levels, as the artifacts to detect cease to exist. Meanwhile, a new AI lab called Deft announced its model, claiming that 86% of user queries are currently judged as fully human by Pangram. The model is in public beta; it notes that short prompts are more likely to be detected as AI.

Related event: Deft Launches Beta Writing Model, Claims 86% of Outputs Pass as Human-Written(6 posts)→

Original post →

More from Models

Models channel →