Internet approaching tipping point of dead internet theory, warns Pangram CEO

Pangram closed a $9 million round for its AI-powered content detection system and announced a partnership with Substack. The integration will let Substack readers see which newsletter authors rely on generative models. At the same time the company launched an AI image-detection tool that extends its existing text-based product. The details were disclosed by CEO and co-founder Max Spero on the Equity podcast of TechCrunch, hosted by Rebecca Bellan.

Spero says the web is “dangerously close” to realizing the dead internet theory within a few years, a scenario where most consumed content is generated by bots for bots and humans become a negligible minority. He adds that synthetic text and images are no longer confined to social-media feeds; they are appearing in résumés, product reviews and insurance claims, leaving platforms and users with basic uncertainty about what is real.

Pangram’s approach does not rely on a binary label of “human or machine” but on a quantitative measurement of the percentage of content produced by a model. Spero argues that this method is technically harder but far more useful because it distinguishes between spot assistance (editing, summarising) and full generation. He notes that false positives carry heightened risk for sensitive images, where misclassifying a genuine photo as fake could cause immediate legal and personal harm.

On the labour side, Spero estimates that basic writing jobs—product descriptions, abstracts, generic SEO copy—will not return. In contrast, high-quality human writing with a distinct voice and original insight may become more expensive as a scarce commodity. He observes that models are closing the gap in “adequacy” but still struggle to create added value that is not derived from existing information.

Embedding the tool in Substack is the first such rollout on a major publishing platform: readers will see an explicit indicator next to the author’s name instead of relying on blind trust. No official benchmarks for the new image version have been released, and the company has not disclosed recall or precision figures for the text version against competitors such as GPTZero or Originality.ai—a point worth remembering before deploying the technology in production.