Pangram becomes publishing's default AI detector on $13M raised

The AI-detection startup Pangram, a 24-person operation headquartered above a Popeyes in Brooklyn, has raised $13 million and become the tool publishers reach for when they suspect a manuscript was written by a machine.

The most visible case is the horror novel "Shy Girl" by Mia Ballard. Hachette acquired it, released it in the United Kingdom in November 2025, and then canceled the US edition after Pangram's cofounder posted on X that the full text scanned as 78 percent AI-generated. Ballard denied writing with AI and said a freelance editor had introduced the changes. It was, per the reporting, "the first known instance of a major publisher walking back a book because of accusations of AI use." A New York Times Modern Love column was later scored 100 percent; a thriller that had sold for $2.4 million was flagged at 97 percent.

Pangram's headline number is a false positive rate of roughly 1 in 10,000 for human text flagged as AI, or 99.98 percent accuracy. The company points to third-party evaluation by researchers at the University of Chicago and the University of Maryland rather than only its own benchmarks. Independent work partly backs it: a study from researchers at the University of Houston and UC Berkeley found Pangram was "the most accurate detector tested," with 97 percent accuracy before back-translation attacks and 94 percent after, at a false positive rate at or below 1 percent.

The problem is what happens when you multiply. Arvind Narayanan wrote on X that the 1-in-10,000 claim should be taken at face value and then run through the math of population-scale deployment. When Vanderbilt disabled Turnitin's AI detector in August 2023, it cited exactly that math: even at a claimed 1 percent false-positive rate, "roughly 75,000 annual submissions would mean about 750 students potentially flagged by mistake."

Pangram was cofounded in 2023 by CEO Max Spero and CTO Bradley Emi, previously on the core computer vision team at Tesla Autopilot. Two of the AI experts we track shared the Wired piece within a day of it running.