New York-based AI detection startup Pangram has raised $9 million in a funding round led by Menlo Ventures, with participation from Haystack, ScOp, Script Capital, and Cadenza. The investment arrives as the company launches its next-generation AI text detection model, Pangram 4, and an AI image detection model, Pangram Image, in a bet that demand for tools distinguishing human-generated content from AI-generated text will continue to grow.
How Pangram’s detection technology works
Pangram’s detection system is built on a large machine learning model trained on tens of millions of known human documents. The startup then creates a “synthetic mirror” for each document, replicating the topic, length, and tone of voice, but written by a frontier LLM. Co-founder Max Spero, a Stanford AI and machine learning graduate, told TechCrunch that the model learns the stylistic differences and choices that AI makes consistently, allowing it to identify AI-generated content with high confidence without relying on metadata or hidden watermarks.
The new text detection model is designed to identify AI-assisted writing and mixed human-AI content, and can also detect content generated by AI humanizer programs that attempt to evade detection. Pangram’s image detection model, currently available via research preview, works on pixel-level distributions, learning subtle statistical differences between real photos and AI-generated images across different AI models, unlike watermark-based systems that only detect output from specific providers.
Funding signals growing demand for AI detection
The $9 million fundraise comes as AI-generated content continues to proliferate across the internet, from SEO slop articles to disinformation campaigns. Spero cited concerns about “LLM-powered Russian disinformation campaigns and UAE-influenced campaigns on Twitter” as driving the need for reliable detection tools. The startup’s emergence coincides with institutional pushback against unvetted AI use: the open-access archive arXiv introduced a policy this year that can trigger a one-year submission ban for authors who fail to review LLM output, including hallucinated references or meta-commentary like “Would you like me to make any changes?”
Also read: Microsoft's Nadella tells enterprises: Don't trust OpenAI or Anthropic too much
Pangram faces competition from other AI detection startups including Winston AI, Originality.ai, Copyleaks, and GPTZero, each building their own detectors. The company offers its technology through a $20-per-month subscription, a Chrome extension that labels posts on platforms including X, LinkedIn, Substack, Reddit, and Medium, and an API. Notable customers include Substack, which integrated Pangram’s technology to show readers which authors use AI, as well as Quora, schools, publishers, and recruiters.
Testing Pangram’s accuracy claims
Independent testing of Pangram’s text detection model showed impressive but imperfect results. The model easily flagged entirely AI-generated news articles from both ChatGPT and Claude, and was rarely fooled by attempts to edit AI-generated text to sound more human. However, it did flag some human-written sentences as AI-generated, particularly in dry, factual news writing that can resemble AI output. In one test, an article that received a 100% human score when submitted as originally written received a 13% AI-assisted score after ChatGPT and Claude were asked to polish it.
The image detection model performed well in testing, easily identifying AI-generated imagery across photorealistic and cartoonish styles. It could also detect AI-generated images appearing within real-world photos, though it incorrectly labeled one photo of an AI-generated image as human content. Spero said roughly one in 10,000 human documents are incorrectly labeled as AI with the current model.
Spero emphasized that the goal is not to fuel a witch hunt against AI users but to provide a mechanism to push back against what he calls “slop.” As he put it: “If we do not actively discriminate in favor of human content, then we’re just gonna get more and more AI, and it’s just gonna drown out any human signal that we have.”
