AI detection startup Pangram Labs raised $9 million, led by Menlo Ventures, to combat the spread of AI-generated content. The funding supports the launch of its new Pangram 4 text detection model, claimed to be over 99% accurate, and a new AI image detector. The move comes as the market seeks reliable tools amid criticism of existing detectors for high error rates and bias.
Pangram funding round
- ▪The funding round was led by Menlo Ventures, with participation from Haystack, ScOp, Script Capital, and Cadenza
- ▪The new funding brings Pangram Labs' total raised capital to almost $13 million
- ▪Pangram Labs previously raised $2.7 million in June 2025
- ▪Pangram Labs, an AI detection startup, raised $9 million in a new funding round
AI detection technology
- ▪Pangram was co-founded by Stanford AI graduates Max Spero and Bradley Emi about two years ago
- ▪The model identifies AI content by learning stylistic differences, not by relying on watermarks or metadata
- ▪Pangram's system is a machine learning model trained on human documents and their AI-generated "synthetic mirror" versions
- ▪The open-access archive arXiv can issue a one-year submission ban for authors who fail to review LLM output
Customer adoption patterns
- ▪Pangram is available to users via a $20-per-month web subscription and a Chrome extension
- ▪Substack integrated Pangram's technology to show readers which authors use AI in their newsletters
- ▪The company's API is used by customers including Quora, schools, universities, publishers, and recruiters
Performance testing results
- ▪Pangram claims a false positive rate of 1 in 10,000 documents, with internal benchmarks for Pangram 4 showing a rate of 1 in 24,000
- ▪A TechCrunch test found the model was effective but sometimes mislabeled human-written sentences as AI-assisted
- ▪Some universities, like the University of Waterloo, have discontinued using AI detectors due to concerns about unreliability and high error rates
- ▪Pangram's new text detection model, Pangram 4, is claimed to be over 99% accurate at finding AI-assisted writing
Image detection model
- ▪The image model analyzes pixel-level distributions to detect images from providers like Midjourney, FLUX, and models from OpenAI and Google
- ▪Research indicates human ability to distinguish AI images from real photos is falling to around 50%, effectively a coin toss
- ▪Pangram has launched an AI image detection model, Pangram Image, which is initially available as a research preview
- ▪The model can reportedly detect an AI-generated image that appears inside a real-world photograph
Story comments
Loading comments…