#How Can Pangram Labs Distinguish Between Human and AI Writing?
Pangram Labs, a startup based in Brooklyn, has developed a detection tool that can differentiate human writing from machine-generated content with remarkable accuracy. The company asserts its technology can identify AI-generated text with a precision exceeding 99.98%. This impressive capability comes with a low false positive rate of merely 1 in 10,000 for human-written material.
Founded in 2023 by former researchers from Stanford University, Pangram is at the forefront of a critical discussion in digital media: the reliability of proving whether a piece of writing originated from a human.
#What Does Pangram Actually Do?
The functionality of Pangram is straightforward. Users submit a piece of writing ranging from 75 words to 75,000 characters, and the tool analyzes it. It classifies the text as Human-Written, Lightly AI-Assisted, Moderately AI-Assisted, or Fully AI-Generated. Additionally, it provides a numerical score reflecting the level of AI assistance, allowing for more detailed analysis.
This innovative detection tool offers support for more than 20 languages and integrates with various learning management systems such as Canvas and Google Classroom, making it versatile and accessible.
A sharper way to see the markets in just 5 minutes.
Same news, different lens. We cut through the noise and hand you the overlooked ideas and the deeper read the crowd misses. Join 38,000+ investors seeing the markets differently.
#How Accurate is Pangram’s Detection Tool?
Independent evaluations from institutions like the University of Chicago Booth School of Business and the University of Maryland have affirmed Pangram’s accuracy. These studies report accuracy rates varying from 99.8% to 100% for content produced by significant language models, including platforms like ChatGPT and Claude. On the RAID benchmark, which is a critical assessment tool for detection methods, Pangram achieved an accuracy of 99.44% along with a mere 0.05% false positive rate.
On January 16, 2026, Pangram also introduced version 3.1, which enhances its capability to analyze mixed-text scenarios where human and AI outputs coexist. This update includes support for multiple languages and tailored detection models specific to various AI systems.
#What is the Current Debate Surrounding Accuracy?
An ongoing academic discourse as of March 2026 highlights a discrepancy between Pangram's advertised false positive rate and some findings from third-party evaluations. While the company boasts a false positive rate of 1 in 10,000, certain external studies have indicated rates ranging from 0.48% to 2%.
Pangram's results in testing varied formats, including academic, news, and creative writing, demonstrated false positive rates between 0% and 0.17%. The essential question emerges: do the controlled settings of these benchmarks accurately reflect the complexities of real-world writing?
Understanding these factors is crucial as the implications of distinguishing between human and AI-generated content have significant consequences for educational integrity, content authenticity, and media consumption.