AI Detection Fails (Again)

AI detection software does not work. Full stop.

A new study from academic researchers shows (again) that these tools are extremely unreliable, and when simple anti-detection techniques are deployed their limited accuracy collapses.

Which means if these tools catch AI cheating, it's bad AI cheating.

While education institutions have relied on plaigarism detection tools as a response to internet research and Wikipedia, it makes sense that we would look for a similar tool for GenAI.

But these tools are just not fit for purpose and can actively cause harm (see the UNG student on academic probation for using Grammarly).

The way forward is not these detectors, but shifting our conversations about academic integrity, so we can have real conversations about how assessment has to change in the age of GenAI.

You can read more about the study here.

Or watch our video from last year, where I trick an AI detector: