What if the tool meant to save creativity from AI actually became its most dangerous gatekeeper? That’s the paradox at the heart of Pangram’s rise—a startup with a mere 24 employees and $13 million in funding, yet somehow positioned as the final arbiter of originality in publishing. It’s not just about detecting AI-generated text anymore; it’s about who gets to decide what counts as ‘real’ writing, and who pays the price for being wrong.
Let’s start with the elephant in the room: Pangram didn’t become a household name by accident. Its sudden infamy came from calling out Shy Girl, a novel that went from being set for traditional publication to being canceled after a 78% AI score. But here’s what most people miss—Pangram’s CEO, Max Spero, admitted he got the manuscript from a pirated site. He didn’t even read the whole thing before running it through his tool. That’s not just sloppy; it’s emblematic of a larger problem. If the person tasked with policing AI is willing to bypass ethical boundaries to make a point, how can anyone trust the system they’ve built?
And then there’s the bias issue. Studies show AI detectors disproportionately flag non-native English writers and neurodiverse authors. Pangram claims it’s better than others, but when three of its most high-profile scandals involved writers of color, it’s hard not to wonder if the tool isn’t just flawed—it’s weaponized. Regina Brooks, president of the Association of American Literary Agents, isn’t wrong to say we need to pay attention to these inequities. After all, who gets scanned, and who gets publicly shamed, is a human decision, not an algorithmic one. What makes this particularly fascinating is how Pangram’s defenders, like computer scientist Tuhin Chakrabarty, argue that the tool shouldn’t be the sole judge. But isn’t that exactly what it’s becoming? A de facto authority in an industry that prides itself on subjective genius.
Spero’s team is small, but their ambitions are vast. They want to be the go-to arbiter for publishers, schools, and even recruiters. Yet their model struggles with short texts and can’t always distinguish between AI editing and human rewriting. The company boasts a 0.0041% false positive rate, but that’s a statistical game. When a writer’s career hinges on a 0.0041% chance of being wrongly accused, the stakes feel absurdly high. I can’t help but think of the authors who’ve been quietly dropped by agents without public fanfare—how many of them were victims of a tool that’s still learning?
And what about the irony that Pangram itself uses AI to train its models? Their ‘synthetic mirroring’ technique involves letting LLMs rewrite human text to teach the system what AI-generated content looks like. It’s a clever trick, but it also means their tool is constantly playing catch-up with the very technology it’s meant to detect. As Notre Dame’s study shows, even a simple ‘humanizer’ tool can trick Pangram into missing AI content 96% of the time. If the system can be outsmarted so easily, what does that say about its reliability? It’s not just about accuracy anymore—it’s about power. Who controls the narrative about what’s ‘real’ versus ‘fake’ in writing?
Spero’s personal demeanor during the interview was telling. He was cagey, distracted, and seemed more interested in managing his public image than engaging deeply with the questions. That’s not just unprofessional—it’s suspicious. When a CEO is too eager to deflect, it raises red flags. And yet, he’s pushing for greater granularity in detection, claiming they’ll soon flag even light AI editing. If that’s true, we’re entering a dystopia where every word is under scrutiny. What happens when a writer’s voice is deemed ‘too AI-like’ because they used a thesaurus? Or when a student’s essay is rejected for being ‘overly polished’ by an algorithm? The line between creativity and plagiarism is getting blurrier by the day.
Ultimately, Pangram’s story isn’t just about AI detection—it’s about trust. Trust in a tool that’s still learning, in a CEO who seems more interested in reputation management than transparency, and in an industry that’s desperate to cling to outdated notions of authorship. As I’ve written before, the real danger isn’t AI itself—it’s the systems we build to control it. And right now, Pangram is both the sheriff and the outlaw in this wild west of creativity. Whether that makes them a savior or a threat depends on who you ask. But one thing is certain: the next time you read a book, you might want to ask yourself—was it written by a human, or by the very tool meant to protect human writing?
