We Built Detectors That Don't Detect Anything

The latest scandal to shake the literary world didn’t arrive with fanfare—it snuck in quietly, the way most institutional crises do these days. Another prestigious prize, another round of AI-generated submissions, another moment where the gatekeepers of culture realized they have no actual gate anymore.

This time it was the 2026 Commonwealth Short Story prize that found itself caught in the familiar bind: works created by artificial intelligence slipping past human readers and judges, only to be unmasked after the fact. The discovery sparked the inevitable scramble for solutions, and like clockwork, the tech world offered what it always offers—a tool. Enter Pangram and its cousins, detector software promising 99% accuracy in identifying machine-written prose.

Here’s the problem: that promise is worthless the moment someone figures out how to break it. Host Kate Lindsay and Ed Zitron, who runs the podcast Better Offline and has built a reputation as a committed skeptic of AI hype, dig into this illusion on a new episode. The real issue isn’t whether we can spot AI writing. It’s that the moment we can’t be sure—the moment there’s even a seed of doubt—institutions have an out. They can shrug. They can say the detection failed. They can move forward without accountability, because now the standard of proof has shifted from human judgment to algorithm output.

This is the trap. These detection tools don’t save us. They provide cover. They give legitimate-sounding reasons to doubt our own ability to know what’s real anymore. And for institutions terrified of being seen as gatekeepers in an era that’s suspicious of gatekeeping, that uncertainty is exactly what they want.