Pangram Raises $9M to Scale AI Content Detection

New York-based Pangram raised $9 million to scale software that distinguishes human-generated content from AI-generated text. The round was led by Menlo Ventures, with participation from Haystack, ScOp, Script Capital, and Cadenza. The fundraise lands as the startup launches Pangram 4, its next-generation AI text detection model, alongside an AI image detection model called Pangram Image.

Pangram says Pangram 4 is over 99% accurate at finding AI-assisted writing and mixed human-AI content, and that it can more easily detect AI humanizer programs. The company’s system was trained on tens of millions of known human documents and synthetic counterparts written by frontier LLMs. Pangram says the detector looks for stylistic differences rather than relying on copy-paste metadata or hidden watermarks.

The article frames Pangram’s growth around a broader shift: AI-generated content is increasingly common across the internet, academic submissions, legal work, and social platforms. TechCrunch cites examples where AI use can lead to ridicule, sanctions, fines, or institutional enforcement. Pangram is also not alone in the market, with Winston AI, Originality.ai, Copyleaks, and GPTZero named as competitors chasing the same demand.

Pangram is available through a $20-per-month web subscription and a Chrome extension that labels posts in real time on X, LinkedIn, Substack, Reddit, and Medium. The extension also provides a feed health score with a human-versus-AI content breakdown. Pangram also offers API access, and TechCrunch reports that Substack recently integrated Pangram’s technology, with other API customers including Quora, schools and universities, publishers and agents, and recruiters.

Pangram Image is currently available in research preview, with a broader release planned in the coming weeks. The system is described as detecting AI-generated images across AI models by analyzing pixel-level distributions rather than relying on watermark-based checks. TechCrunch’s testing found the image detector impressive but not perfect, including one instance where it incorrectly labeled a photo of an AI-generated image as human content.