Pangram just put real money behind a hard question: can you actually tell when text or an image was made by AI? According to TechCrunch AI, the startup raised $9 million for its AI detection system and signed a partnership with Substack, which is now using Pangram’s tech to flag which newsletter authors lean on AI to write. TechCrunch AI reports that co-founder and CEO Max Spero laid out the details on the Equity podcast, and the short version is this: detection is a lot messier than a simple ‘real or fake’ verdict.
📌 The news in brief
- Pangram closed $9 million in funding for its detection platform.
- The company struck a deal with Substack, giving readers a signal about whether their favorite writers use AI.
- It shipped a new AI image detection tool, pushing beyond text into visual content.
That combination matters. A funding round is one thing. A funding round plus a live partnership with a major publishing platform plus a product expansion is a company moving fast on all three fronts at once.
🌐 Why this matters
The internet has a trust problem, and it runs deeper than social feeds clogged with AI slop. As TechCrunch AI notes, AI-generated text and images are showing up in job applications, product reviews, and even insurance claims. When you can’t tell what a human actually wrote or shot, every one of those systems gets easier to game.
That’s the gap a wave of startups is racing to fill. They want to be the ‘trust layer’ the web never had. Pangram is one of the loudest names in that group right now, and the Substack deal gives it something most competitors lack: distribution and a real-world test at scale.
🎯 The harder problem Spero raised
The most useful takeaway from Spero isn’t the money. It’s the line he keeps pointing at: the difference between AI assisted and AI generated. Those aren’t the same thing, and that distinction breaks a lot of naive detection.
- A writer who drafts every word, then runs it through a grammar tool, is AI assisted.
- A writer who types a prompt and publishes the output with light edits is closer to AI generated.
- Most real work now lives in the middle, which is exactly where a blunt ‘real or fake’ label falls apart.
What stands out here is that Pangram is selling nuance, not a lie detector. That’s a smarter pitch, because the market has already learned that overconfident detectors produce false accusations, and false accusations are how these tools lose trust with the people using them.
🔭 What to watch next
- False positives. Any detector aimed at real authors will eventually flag a human as a machine. How Pangram handles disputes will decide whether writers trust the Substack label or resent it.
- The image tool. Text detection is hard. Image detection, as generators keep improving, may be harder. This is the area to stress test.
- Platform adoption. Substack is the first big name. If hiring platforms, review sites, or insurers follow, detection stops being a novelty and starts being infrastructure.
- The arms race. Every detection gain invites a new evasion trick. Expect this to stay a moving target, not a solved problem.
💬 My take
Detection tools are having a moment because the pain is finally concrete. It’s no longer an abstract worry about slop. It’s your hiring pipeline, your review section, your claims desk. Pangram betting on the assisted-versus-generated line is the right instinct, because that gray zone is where trust actually gets decided.
The open question is whether any detector can stay accurate as the models it hunts keep getting better. That tension is the whole story, and it isn’t going away soon.
For the full conversation with Max Spero, including where he draws the line on acceptable AI use, check the original Equity episode covered by TechCrunch AI.