The flood of AI-generated text into places where humans used to be the default — job applications, product reviews, insurance claims — has created a mess that platforms are only now starting to clean up. Pangram, a startup that’s made AI detection its entire pitch, just raised $9 million and landed a notable customer: Substack. The newsletter platform is now using Pangram’s technology to show readers which of their favorite authors are leaning on AI to write.
Pangram co-founder and CEO Max Spero sat down with TechCrunch’s Equity podcast to talk through the promise of detection tools and where the line sits between AI-assisted and AI-generated content. That distinction matters more than most people admit. Using AI to brainstorm or clean up grammar is fundamentally different from hitting generate and publishing whatever comes out. Spero’s framing suggests the company is thinking carefully about that gray zone rather than just slapping a binary ‘AI or not’ label on everything.
The startup also recently shipped an AI image detection tool, extending its reach beyond text. That timing tracks with the broader arms race unfolding across the internet. Google, OpenAI, and others have all promised watermarking or detection capabilities, but independent verification layers like Pangram are emerging because the platforms that generate AI content have a conflicted interest in policing it. A third-party detector that doesn’t also sell the generative model has a cleaner incentive structure.
Whether readers actually want an ‘AI score’ attached to every newsletter is an open question. Transparency could push some writers to disclose their process voluntarily, while others might simply migrate to platforms that don’t nudge them. The $9 million in fresh funding suggests investors see demand, but the real test is whether detection accuracy holds up as language models improve. False positives — flagging a human writer as AI — would be a reputation killer for any platform that adopts this too aggressively.
