Caden Qiao was a junior at a state school in the Midwest when he posted a small piece of code online last spring.
It scanned essays for the telltale fingerprints of AI writing.
Within weeks, professors were forwarding it to colleagues.
Within months, he says, two companies sent him letters threatening legal action.
Qiao is not a lawyer or a privacy researcher.
He is a college student with a laptop and a stubborn belief that something had quietly broken in American classrooms.
Since 2023, a large share of written work submitted by students has been at least partly machine-generated.
Almost no one can prove it in a way that survives a formal appeal, because the tools that detect AI are themselves uncertain, and the companies selling those tools rarely promise accuracy.
Into that vacuum walked a twenty-year-old with a free script.
Human writing has a lumpy cadence, odd word choices, digressions that go nowhere.
Machine writing tends toward a smooth, frictionless sameness.
He built a scoring model around that difference and released it without a paywall.
According to screenshots he shared with reporters, campus download traffic spiked during finals week.
A plagiarism-detection firm whose product Qiao's tool frequently contradicted sent a cease-and-desist.
A university's IT office quietly blocked his domain, citing "network policy." On a Reddit thread that ran to 900 comments, a graduate instructor wrote that she had stopped assigning take-home essays entirely, because she could no longer tell who was doing the work.
We built a system where a student's grade depends on a written artifact, and then we handed everyone a machine that produces written artifacts instantly.
Now we are arguing about detection instead of asking whether the assignment still means anything.
The defenders of the AI tools argue that detection is impossible and that chasing it wastes time.
Statistical detectors mislabel human writing, especially the writing of non-native English speakers, at rates that should embarrass anyone who uses them as evidence.
But "we cannot detect it perfectly" has become a slogan for "we should stop trying," and that is a different claim.
Qiao's real sin, in the eyes of the industry, is that he made the problem visible for free.
A detection product sold by a vendor can be managed, priced, and quietly retired when it fails.
A script passed around by students cannot.
He turned an expensive institutional headache into a public one.
He told one interviewer that he expected to be ignored, not threatened.
He says the letters taught him more about institutional self-protection than any course he has taken.
There is a version of this story where Qiao is a hero and a version where he is a nuisance.
A system that cannot verify its own credentials will eventually stop issuing them credibly, and the people harmed most will not be the ones gaming it.
They will be the students who did the reading, wrote the essay, and watched a classmate skate by on autopilot. **Closing thought:** We spent a decade telling young people that writing was a skill worth having, then flooded the market with a machine that makes the skill optional and the shortcut invisible.
Final Thoughts
Until institutions decide what they actually want to measure, every detector, paid or free, is just a thermometer arguing with a fever.