Brooklyn AI Firm Pangram Claims to Spot Machine-Written Text Amid Publishing Turmoil

Pangram, a Brooklyn-based AI detection startup with 24 staff, has secured $13 million in funding and introduced its latest model, Pangram 4, which estimates the share of AI in a text. After the firm flagged Mia Ballard's self-published novel as 78 percent AI-generated, Hachette cancelled the deal, and similar accusations followed for a New York Times column, a Commonwealth Short Story Prize winner, and a multimillion-dollar thriller.

In July, Substack announced integration of Pangram so users can scan content for possible AI use. The company describes its approach as "synthetic mirroring" and "hard negative mining," claiming a false-positive rate of 0.0041 percent, though scholars warn the tool can misidentify non-native and neurodiverse writers. Major publishers either decline comment or treat the detector as a supplemental aid, while some literary agents employ it to confront authors about AI involvement. Pangram’s leadership acknowledges controversy, offers cash rewards for proven false flags, and plans to add finer-grained detection for light AI editing.