Pangram Secures $9 Million to Combat AI-Generated Content

Founders Max Spero and Bradley Emi launched Pangram two years ago, motivated by the proliferation of automated SEO content and disinformation campaigns. The company's core system functions as a large machine learning model trained on tens of millions of human documents. By creating a synthetic mirror of these texts, the software identifies consistent stylistic choices and patterns unique to frontier large language models. Rather than relying on metadata or watermarks, the engine analyzes the underlying structure of the writing.

The demand for such verification is moving beyond individual concern into institutional policy. Organizations like arXiv have begun issuing submission bans for researchers who fail to review AI-generated hallucinations, while legal professionals face sanctions for using fabricated citations. Pangram’s tools, available via a web subscription or browser extension, now provide real-time labels on platforms like X, LinkedIn, and Substack. Substack has already integrated the technology to offer readers transparency regarding AI-assisted newsletters. While testing shows the model is highly effective, it remains a work in progress, occasionally misidentifying human-written prose as AI-assisted. Spero maintains that the goal is not to eliminate AI usage, but to establish a necessary mechanism for transparency as synthetic content threatens to overwhelm human-generated information.