Pangram, an AI detection startup, announced a $9 million funding round and a new detection model that it says can accurately identify AI-generated text 9,999 times out of 10,000. The announcement comes as generative AI floods the internet with synthetic content, forcing publishers and platforms to grapple with how to distinguish human work from machine output.
Max Spero, the company's chief executive, calls himself the internet's slop janitor. The New York-based startup has already signed up high-profile online publishers as customers, offering a tool to sniff out AI-written articles, social media posts, and now images.
The internet's slop janitor
Spero and co-founder Bradley Emi, both 30, met as undergraduates at Stanford's artificial intelligence lab. After graduating, Spero worked at Google and Emi at Tesla, where he worked under Andrej Karpathy, the prominent AI researcher and OpenAI co-founder. The release of ChatGPT in 2022 spurred them to build something "for the world that this technology will bring forth," Spero said.
A new model and fresh funding
The $9 million round was led by Menlo Ventures, with participation from ScOp Venture Capital, Haystack Ventures and others. Alongside the funding, Pangram released Pangram 4.0, its latest text-detection model, and its first-ever AI-image detection model. The company claims the text model correctly flags AI content in 9,999 out of 10,000 cases.
The thorny questions of AI detection
AI writing and detection tools raise difficult questions. Even with near-perfect accuracy, a tool that is wrong once in 10,000 tries could still generate false positives that damage a writer's reputation or a publisher's credibility. The broader debate touches on how society should treat AI-generated text and what role detection should play.
For professionals working with Generative AI and LLM systems, the reliability of such tools is a practical concern. Pangram's high-profile publisher customers suggest that some media organizations are already integrating detection into their workflows. But the company acknowledges that even rare errors can have consequences, especially if detection is used to make automated decisions about content.
Why this matters for writers and developers
The arrival of tools like Pangram adds a new layer to the conversation around AI for Writers, as detection capabilities could influence how writers use AI assistants. For writers, the tool's near-perfect accuracy could offer a way to verify human authorship, but the risk of false positives means that any automated flag should be reviewed by a person. IT and development teams building content pipelines may need to incorporate detection APIs while building in human oversight. The technology's integration into publishing could become a standard feature, much like plagiarism checkers today.
Your membership also unlocks:
