New York-based startup Pangram has raised $9 million to expand its AI content detection technology, betting that demand for tools distinguishing human-written from AI-generated material will keep growing as "AI slop" spreads across the internet. The funding round, led by Menlo Ventures with participation from Haystack, ScOp, Script Capital and Cadenza, coincides with the launch of Pangram 4, a next-generation text detection model, and Pangram Image, a new tool for spotting AI-generated images.

Pangram claims its new text model is over 99% accurate at identifying AI-assisted and mixed human-AI writing, and is better at catching so-called "AI humaniser" programs designed to disguise machine-written text. Founded roughly two years ago by Stanford graduates Max Spero and Bradley Emi in response to the flood of bots and disinformation following ChatGPT's launch, the company trains its models by comparing genuine human documents against AI-written "synthetic mirrors" of the same topic and tone, rather than relying on metadata or watermarks. The image detector remains in research preview for now. Pangram faces competition from rivals such as Winston AI, Originality.ai, Copyleaks and GPTZero, as institutions including arXiv introduce policies penalising unreviewed AI-generated submissions.