The rise of AI detection tools like Pangram has sparked a cultural reckoning in the world of writing—one that feels less like a technological breakthrough and more like a moral tightrope walk. Here’s the thing: we’re not just talking about software that flags AI-generated text. We’re talking about a tool that’s now shaping careers, dictating which stories get published, and deciding who gets to claim authorship in an era where the line between human and machine is blurring faster than anyone anticipated. And yet, the more I think about it, the more I wonder: is Pangram really the hero of this story, or is it just another algorithmic gatekeeper with its own biases and blind spots?
Let’s start with the basics. Pangram, a Brooklyn-based startup with a budget that pales in comparison to OpenAI, has somehow become the go-to arbiter of originality. Its CEO, Max Spero, is a self-described ‘AI police’ who’s taken it upon himself to expose literary scandals. But here’s what’s fascinating: the company’s rise to prominence was fueled by a single scandal—the accusation that a novel titled Shy Girl was 78% AI-generated. Hachette canceled the book’s release, and suddenly, Pangram was thrust into the spotlight. What makes this particularly fascinating is how quickly the public seems to have embraced this tool as an infallible judge of creativity, even though its own methods are as opaque as the AI it claims to detect.
Spero’s story is almost too convenient for a Silicon Valley origin myth. A Stanford dropout who worked at Google and Nuro, he founded Checkfor.ai (now Pangram) after seeing the writing on the wall: AI was going to flood the content landscape, and someone had to clean up the mess. But here’s the rub: the very technology that Pangram uses to detect AI—‘synthetic mirroring’ and ‘hard negative mining’—is itself a product of AI. It’s like using a mirror to catch a reflection of a mirror. What does that say about our trust in these tools? Are we just outsourcing our judgment to something we barely understand? In my opinion, this is where the real danger lies. We’re so focused on the threat of AI-generated content that we’re ignoring the fact that the tools we use to combat it are equally untrustworthy.
Then there’s the issue of bias. Critics argue that Pangram’s algorithms disproportionately flag non-native English writers and neurodiverse authors, effectively weaponizing the very systems that claim to promote fairness. Sam Illingworth, a professor of critical AI literacy, isn’t alone in pointing out that detectors like Pangram are ‘prejudiced against certain people.’ This isn’t just a technical flaw—it’s a societal one. If AI detection tools are being used to gatekeep literary success, who decides who gets to be ‘human enough’ to publish? And what happens when the algorithm’s definition of ‘originality’ is shaped by the same cultural norms that have historically excluded marginalized voices? From my perspective, this isn’t just about accuracy; it’s about power. Who controls the narrative, and who gets to define what counts as ‘real’ creativity?
The controversy surrounding Shy Girl also raises a deeper question: can a single percentage truly capture the essence of a story? Spero’s team analyzed the manuscript, but they didn’t read it. They didn’t consider the author’s intent, the cultural context, or the emotional resonance of the work. They just ran it through an algorithm and called it 78% AI. What this really suggests is that we’re reducing the complexity of human expression to a number—a number that can be manipulated, misinterpreted, or weaponized. And yet, we’re treating this number as if it’s the final word. How absurd is that? If you take a step back and think about it, this is the same mindset that led us to trust Facebook’s algorithms to curate our news feeds, or Google’s search results to shape our understanding of the world. We’ve outsourced our judgment to systems that were never designed to understand nuance.
Pangram’s defenders argue that the tool is just one part of a broader editorial process. But here’s the thing: in an industry where a single accusation can derail a career, that ‘one part’ is often the only part that matters. When a publisher sees a 100% AI score on a submission, they don’t need to dig deeper—they just reject it. The same goes for agents, editors, and even readers who now use Substack’s integration to scan articles before they read them. This raises a terrifying possibility: what if the real enemy isn’t AI, but the fear of it? What if the rush to adopt detection tools is less about protecting human creativity and more about avoiding the discomfort of admitting that AI might be capable of producing work that’s indistinguishable from ours? A detail that I find especially interesting is how Pangram’s CEO has been so quick to distance himself from the ethical implications of his product. He talks about transparency and trust, but when asked about the potential for false positives or bias, he deflects. It’s almost like he’s trying to convince himself that this isn’t the same kind of power play that happened in the early days of social media or search engines.
And let’s not forget the irony of it all. Pangram’s success is built on the very AI it claims to detect. Its models are trained on datasets that include human writing, which is then mirrored by LLMs to teach the system what AI-generated text looks like. But what happens when the AI starts to mimic human writing so well that the tool can’t tell the difference? What happens when the algorithm’s training data includes works that were themselves generated by AI? This isn’t just a technical challenge—it’s a philosophical one. If AI can create art that moves us, why should we care if it’s ‘real’? And if we do care, who gets to decide what counts as ‘real’? The answer, it seems, is whoever controls the algorithm.
In the end, Pangram’s story is a microcosm of our relationship with AI: full of promise, but also fraught with contradictions. It’s a tool that could democratize creativity by holding AI accountable, but it’s also a gatekeeper that risks replicating the same biases and exclusions that have plagued the publishing industry for decades. As we move forward, I think we need to ask ourselves a harder question: are we using AI to enhance human creativity, or are we using it to replace it? Because the difference between the two might be the difference between a thriving literary culture and a world where the only stories that matter are the ones that pass a machine’s test.
