Pangram has become a de facto benchmark after publishing scores that led publishers and prize committees to question or withdraw works.
The company’s profile rose sharply when it flagged a self-published novel as 78% AI-generated and later assigned 100% scores to other high-profile pieces, and Substack said it was integrating Pangram into its platform.
Wired’s reporting traces the surge to a string of public calls and media moments that turned a little-known tool into a category influencer.
Pangram detects AI involvement by training models to reproduce human text, a technique it calls synthetic mirroring, and by using what it describes as hard negative mining to hunt for troublesome false positives.
The startup remains small: it lists about 24 employees, has raised $13 million to date, and in July raised $9 million while launching Pangram 4 and advertising six new roles.
Wired records that Pangram emerged as a front-runner after independent testing, a position that has won customers in education and publishing even as many writers react angrily to automated judgments.
"We think they’ve lost a lot of trust, especially in the world of creatives," Pangram CEO Max Spero said, and he added, "All of our datasets are properly licensed, which I think is kind of rare in the AI world today."
