Pangram, a startup focused on AI-detection, recently raised $9 million and entered into a partnership with Substack. Substack is now using Pangram’s technology to indicate to readers which of their preferred writers use AI when composing newsletters. Pangram also launched a new AI image-detection tool.
The interview: Max Spero on the "dead internet" scenario
Max Spero, co-founder and CEO of Pangram, joined Rebecca Bellan on TechCrunch’s Equity podcast to discuss the promise of AI-detection tools and how to distinguish AI-assisted from AI-generated content.
Spero warned that current trends could bring the internet "dangerously close" to the so-called "dead internet" theory within a few years — a scenario in which a large portion of online content is produced by automated systems, fundamentally altering web traffic and trust.
Why a binary label isn't enough
Spero argued that it is often more difficult — and more useful — to measure how much AI contributed to a piece of content than to apply a simple human/AI label. A more granular approach can help platforms and users better assess provenance and reliability.
Risks of false positives
Spero also stressed the stakes of getting AI-detection wrong. False positives can have serious consequences, especially when sensitive images or personal data are mistakenly classified as AI-generated. Such errors can produce legal and ethical issues and erode trust.
Labor market effects: loss of low-tier writing jobs and greater value for quality human writing
According to Pangram’s leadership, many low-end, high-volume writing jobs may disappear permanently as AI handles those tasks. Conversely, genuinely good human writing could become more valuable, as authentic, creative, high-quality content differentiates itself.
Why this matters
As AI-generated content takes up a growing share of online information, accurate detection and transparency become increasingly important for users, platforms, and regulators. Companies like Pangram aim to serve as the internet’s "trust layer," but technical limitations and the consequences of misclassification mean deployment carries risks.
The full discussion is available on the TechCrunch Equity podcast, which explores further arguments and details about the current state and future of AI detection.
