Pangram, a Brooklyn-based AI startup with 24 employees and $13 million in funding, has positioned itself as a leading AI-detection service, often sparking controversy. The company’s tool assigns a percentage score to text, suggesting how much AI may have contributed to its creation. In January, Pangram’s CEO claimed the novel *Shy Girl* was 78% AI-generated, contributing to Hachette’s cancellation of its release. Similar accusations followed, including a Modern Love column in *The New York Times* (100%), a Commonwealth Short Story Prize winner (100%), and the novel *Daggermouth* (60%).
Substack recently integrated Pangram into its platform, allowing readers to check for AI use in posts. The move reflects growing demand to distinguish AI-generated content from human writing, though not all writers support such tools. Critics argue detection software is as problematic as the AI it targets, with some calling it a career-ending risk. Pangram’s cofounder, Max Spero, acknowledges the tool’s role in disputes but downplays its impact, stating that book deals are rarely decided by detection scores alone.
Pangram’s detection relies on synthetic mirroring and hard negative mining, training its model to recognize AI patterns. The company has expanded beyond publishing into education, legal, and recruitment sectors. While some authors and agents use Pangram to vet manuscripts, others warn of false positives and potential biases in its machine learning. Research suggests detectors disproportionately flag writing from non-native English speakers or marginalized groups, raising ethical concerns.
