Detectors Out

Yale's teaching centre says it "doesn't endorse the use of AI detection software". Its Turnitin page is plainer still: "The AI detection feature of Turnitin is currently disabled."

Cornell's teaching centre doesn't expect a fix. Its own page: "Unfortunately, it is unlikely that detection technologies will provide a workable solution to this problem."

Yale and Cornell weren't the first. Vanderbilt switched Turnitin's AI detector off in August 2023: "Vanderbilt has decided to disable Turnitin's AI detection tool for the foreseeable future."

Its reason was its own arithmetic. It sent 75,000 papers to Turnitin in 2022, and by its own estimate "around 750 student papers could have been incorrectly labeled". UW-Madison chose not to start: it "decided not to offer an enterprise tool for AI detection".

Washington State University cancelled its AI-detection contract in February. Its memo to instructors explains why, in its own figures.

In one term, Fall 2024, the detector checked 148,547 assessments. The memo works at a 1% false-positive rate. At that rate, it says, "almost 1,485 assessments were likely flagged" when they weren't AI.

The chart puts the two side by side. The false flags barely leave the axis, and that is the memo's point: a small rate on a big number is still almost fifteen hundred assignments.

Then the cases. Between 2023 and 2025, a third of the AI cases that reached Washington State's hearing boards ended "not responsible", because the detector was the only evidence.

The memo draws the line itself. Suspicion from the detector alone, it says, "is not sufficient".

The 1% is not settled. The memo calls it disputed: "Although disputed, Turnitin claims that they have a false positive detection rate… of 1-2%".

It says "other studies placed this value higher, especially for students who are neurodivergent or have English as a second language". And it adds that "even experts can have false positive detection rates of 4% or more".

Turnitin's own figures from June 2023 were different: "less than 1%" for whole documents with 20% or more AI writing, and a sentence-level rate of "around 4%".

Even the maker says so. Turnitin's own guide says its AI detection "should not be used as the sole basis for adverse actions against a student."

Its report now shows low scores, from 1 to 19%, as an asterisk rather than a number.

And in America's schools? A national survey of more than a thousand public-school teachers and principals asked: "Are there specific policies or guidelines that govern Gen AI use in your school?" Only 18% of teachers said yes. Among principals it was 24%.

The tool that promised to catch AI is being switched off. In America's schools, what's left, for now, is what the researchers call "AI-dar": a teacher's hunch.

That is the researchers' word for schools, not universities. Washington State's memo has already said a hunch from a detector is not enough.

Three things to watch. Whether more universities say in their own words that they have switched detection off; we name five here, and couldn't confirm how many there are in all.

Whether schools' rules catch up with their teachers' dilemmas, when only 18% of teachers say there are any.

And whether Turnitin publishes a current false-positive rate of its own, in place of a 2023 blog and a memo's disputed reading of it.