Pangram Labs: Robust and Reliable AI-Generated Text Detection

I: AI Detection - Overview

The spread of large language models (LLMs) in the past several years has resulted in a significant increase in AI-generated content in many areas of professional activity, academia, and personal life. As a result, a new category of content moderation has emerged - AI detectors. AI detection involves detecting AI generated content and classifying the source of the AI generated content. 

The surge in AI-generated content is reshaping how we assess authenticity online. Students are using ChatGPT to write essays and exams. Fake scholarly articles and “hallucinated” legal citations are infiltrating official submissions. AI-generated resumes, doctored financial documents, and spammy product reviews are slipping through hiring and commerce platforms. From education to enterprise, from media to government, institutions are under pressure to respond. Schools are implementing originality checks. Platforms are adopting AI detection to self-regulate. Governments are mandating transparency. AI detection is emerging as a foundational layer in the modern information stack; vital for upholding trust, enforcing standards, and managing reputational and regulatory risk.

II: Use Cases


1. AI Detection for Teachers


A clear use case for an AI detector is in education. AI detectors for teachers can reliably screen student work for AI content, such as AI generated essays, discussion posts, homework assignments, exams, or other educational material. Traditional plagiarism is being replaced by students leveraging ChatGPT and other LLMs to generate content, and traditional plagiarism checkers must evolve to detect AI for schoolwork as part of their feature set. AI text detectors that focus on classifying text as human-written or AI-generated are critical tools looking to detect AI in student work.

Universities can use an AI detector for screening applications, personal statements, and question responses from applicants for admission. Ensuring authenticity in personal statements and college applications is essential for admissions boards to be confident that they are seeing an honest representation of the applicant and their work. AI detectors for universities will serve a critical role in enabling universities to reliably detect AI generated work from applicants.

2. AI Detection for Publishers

Publishers face a crisis of authenticity due to the rise in AI generated content across books, articles, and other published content. AI writing detectors and AI content detectors are critical tools for publishers to verify genuine content and ensure the legitimacy of their publications. AI detectors for publishers allow publishers to screen their work for potentially inauthentic AI-generated content, and guarantee original work in their output.

3. AI Detection for Content Moderation

An emerging frontier for AI detection is content moderation and social media. Since the spread of LLMs, bot accounts and spammers have increased their reach on social platforms and online forums. Administrators, content moderators, and stakeholders in social media sites can leverage AI detection to identify and classify AI content, weeding out inauthentic material and preserving the authenticity of their sites. AI detection for content moderation and AI detection for social media is an important and emerging frontier for sites preserving authentic online discourse and content

4. AI Detection for Job Applications


HR professionals and hiring managers face a high volume of applicants for positions. The increase of AI usage results in many resumes, questionnaire responses, and cover letters using AI to generate seemingly authentic content. AI detectors for HR give hiring professionals the tools they need to verify legitimacy of job applications, screen for AI, and gain critical visibility into the source of application content.


III: False Positives

A key consideration for use of an AI content detector is minimizing the false positive rate. The false positive rate of an AI detector means the percent of times that the AI detector classifies content as AI-generated when in fact it is authentic content. This is a critical metric, as a high false positive rate makes an AI detector or AI detection algorithm unreliable. For example, with AI detection for education, if a teacher is using an AI detector to evaluate student submissions for AI, just a 1% false positive rate means that with a class 

size of 50 students, a teacher will have an issue of false classification of AI for a student submission once in every two cases.

Pangram Labs is the best AI detector in minimizing false positives, with an approximately 1 in 10,000 false positive rate. This metric is far superior to competitor stated false positive rates, and ensures that users can count on predictions of AI generated content when making content related decisions. Pangram conducted a comprehensive benchmark of 1,976 documents across 10 diverse text domains, including student writing, creative writing, scientific writing, books, encyclopedias, news, email, scientific papers, and short-form Q&A to validate their position. The benchmark also spanned content generated by eight leading closed and open source LLMs, including GPT-3.5, GPT-4, Google Gemini Pro, Mistral 7B, and LLaMA 2. Pangram’s model outperformed all commercial alternatives, achieving over 99% accuracy, surpassing competitors that failed to reach even 95%. Most notably, Pangram’s false positive rate was 38x lower than other tools on average, and 3x lower than the next best solution (GPTZero). This matters: in high-stakes environments like education, publishing, and legal compliance, minimizing false positives is a prerequisite for adoption. Pangram’s performance advantage positions it as a category leader in both precision and reliability.

IV: Pangram Labs

Pangram Labs has created the world’s leading AI detection model, capable of reliably detecting AI generated text across all widely used language models. Pangram’s team is composed of leading researchers in machine learning and artificial intelligence - the result is industry leading reliability and best-in-class accuracy, all while maintaining lower false positive rates than any other AI detectors available today. Pangram Labs meets the need for AI detection across all of the use cases above, and their technical expertise in the area gives them an advantage in detecting AI generated content across models and mediums. Pangram’s free AI detector and comprehensive AI detection products provide a wide range of functionality for those looking to check text for AI.

Pangram’s founders, Max Spero and Bradley Emi, met while getting their master's degrees in artificial intelligence at Stanford. Before launching Pangram, Bradley led the deep learning research group at Absci, a generative AI drug discovery company, and previously was a member of the core computer vision team at Tesla Autopilot. Max previously worked on autonomous vehicles at Nuro, leading their active learning effort. He has a long history of deploying successful machine learning products at Google, Two Sigma, and Yelp.