The dead internet theory — the idea that most online activity is bots talking to bots — isn’t just a paranoid fantasy anymore. Max Spero, co-founder and CEO of AI detection startup Pangram, thinks we’re “dangerously close” to that reality, and he just raised $9 million to do something about it.
Pangram’s pitch is straightforward: be the trust layer for platforms drowning in AI-generated content. Substack is already using the tech to flag AI-written newsletters, and Pangram just launched an image detection tool to catch synthetic visuals too. But Spero’s real insight on TechCrunch’s Equity podcast is more nuanced than “AI bad, human good.” He argues the useful question isn’t binary — AI or not — but how much AI went into something. An author using AI to clean up grammar is different from one letting a model write entire essays, and platforms need tools that can tell the difference.
The stakes get uncomfortable fast. False positives in AI detection could wrongly flag human work, and when sensitive images are involved, the consequences get serious. Spero doesn’t sugarcoat the tradeoffs: detection systems that err on the side of caution risk smearing legitimate creators, while lenient ones let synthetic content flood through. That tension is why Pangram’s approach focuses on calibrated confidence rather than binary labels.
Spero is also blunt about what this means for labor. The “bottom tier” of writing jobs — formulaic product descriptions, basic SEO content — may be gone for good. But he sees a flip side: genuinely good human writing could become more valuable precisely because it’s now verifiably rare. That’s the bet Pangram is making, and Substack’s adoption suggests at least some platforms agree. Whether detection tools can stay ahead of generation models is the open question. The track record of AI detectors is checkered at best — earlier tools from OpenAI and others were quietly shelved after embarrassing accuracy failures. Pangram’s $9 million says Spero thinks his approach is different. The next few years will show whether he’s right, or whether this is another arms race where detection is perpetually one step behind. For now, the fact that Substack is willing to attach labels to its own creators signals a shift: platforms are starting to treat AI disclosure as a feature, not a liability.
