AI detection company Pangram Labs has raised $9 million in a Series A round led by Menlo Ventures, with participation from Haystack, ScOp Venture Capital, Script Capital and Cadenza. The raise brings its total funding to nearly $13 million.
The company builds classifier models, neural networks trained on human text written before 2021 and paired with AI-generated samples, to judge whether content was machine-made. Its newly announced Pangram 4 model claims a false-positive rate of about 0.0041%, or roughly one in 24,000 documents, and adds resistance to humanizer tools designed to disguise AI writing.
Pangram is also expanding into images, with a model it says can flag outputs from systems like GPT-4, Gemini and Midjourney as well as video generators such as Kling AI.
The article is candid about the limits of the field. Industry-wide, detectors remain unreliable: MIT Sloan calls the technology far from foolproof, studies have found bias against non-native English writers, and institutions including MIT and the University of Waterloo have pulled back from using detectors as definitive judgments.
Source: SiliconANGLE.
Comments 0