Pangram’s Max Spero on why AI detection is harder than ‘Real or Fake’
TechCrunch Theresa Loconsolo ● Covered by 2 sources
Pangram says AI detection is still messy, even for text and images. It just raised $9 million and Substack is already using its tools.
Based on reporting by TechCrunch, Theresa Loconsolo — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
The web’s trust problem is getting harder to ignore, and AI-generated content is a big reason why. It’s showing up in places that used to feel at least somewhat human: job applications, product reviews, insurance claims. Once that happens, the whole question shifts from “Is this good?” to “Did a person even make it?”
Pangram is trying to be part of the answer. The startup recently raised $9 million for its AI detection system, and it has already landed a partnership with Substack. That matters because Substack is using Pangram’s technology to show readers which newsletter authors use AI in their writing. It’s a direct attempt to label the gray area instead of pretending it doesn’t exist.
The company has also added an AI image detection tool, which suggests it’s not treating this as a text-only problem. That makes sense. The same incentives that push people to pump out synthetic prose are now pushing synthetic pictures into the same channels, where platforms need some way to sort signal from sludge.
Max Spero, Pangram’s co-founder and CEO, is set to talk through the limits of AI detection on TechCrunch’s Equity podcast, including where the line sits between AI-assisted and AI-generated. That line is the whole fight, really. If detection tools can’t explain that boundary clearly, they risk becoming just another layer of internet theater.
The conversation around AI detection is drifting toward a familiar trap: people want a clean yes-or-no answer in a world that’s full of blur. Pangram’s Substack deal is a sensible move because labels beat vibes, but the industry still loves pretending “trust layer” is a product category instead of a hopeful slogan. That usually ends the same way: with a lot of confidence and not enough proof.
My take — AI-written commentary, not fact-checked reporting
AI detection is useful when it is narrow and honest, and mostly nonsense when marketed like moral polygraphy. The smarter play is transparency, not magical certainty; the internet already has enough black boxes pretending to be referees.
Read more about this at: TechCrunch