AI detection startup Pangram secures $9m to verify human content online

New York-based artificial intelligence 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 capital injection coincides with the launch of its next-generation text detection model, Pangram 4, and a new AI image detection tool.

This investment highlights a growing market imperative for platforms and institutions to verify human authorship. As synthetic text and imagery proliferate across digital spaces, companies and academic bodies are increasingly seeking reliable mechanisms to maintain trust and avoid operational risks.

Pangram claims its new text model is over 99 per cent accurate at identifying AI-assisted writing and mixed human-AI content, including text processed by AI humaniser programs. The accompanying image detector, currently in research preview, analyses pixel-level distributions to spot synthetic imagery across different models rather than relying on proprietary watermarks.

Stanford graduates Max Spero and Bradley Emi founded the company approximately two years ago in response to the surge of bot-generated search engine content and coordinated disinformation campaigns. Spero emphasised that knowing the origin of a text fundamentally changes how readers approach it, particularly when assessing potential hallucinations or journalistic integrity.

The detection system operates by training a large machine learning model on tens of millions of known human documents. It then generates a synthetic mirror of each document using a frontier large language model to learn the consistent stylistic choices made by artificial intelligence.

Institutional verification

The demand for such verification is being driven by tangible institutional backlash against unchecked AI usage. The open-access archive arXiv recently introduced a policy imposing a one-year submission ban on authors who fail to review large language model output, such as hallucinated references.

Pangram monetises its technology through a $20-per-month web subscription, a Chrome extension, and enterprise application programming interfaces. Notable API clients include Quora, various educational institutions, publishers, and recruiters, while Substack has integrated the technology to disclose AI usage by newsletter authors.

Independent testing indicates the system is highly effective at flagging entirely synthetic articles and resisting basic humanisation prompts. However, it occasionally misidentifies heavily edited human text as AI-assisted. Furthermore, the image detector once incorrectly classified a photograph of an AI-generated image as human content.

Spero warned that the sheer volume of artificial output threatens to overwhelm digital spaces. "We are getting new GPUs faster than new people are being born," he said, arguing that active discrimination in favour of human content is necessary to prevent the human signal from being drowned out.