Nearly half the students in one of Purdue's biggest programming classes got flagged for using AI on their homework.
A team of researchers built Argus, a detection system that watches how students write C code over the course of an assignment, not just the finished product. It tracks behavioral and stylistic signals, like how code evolves, to flag work that looks LLM-assisted. Applied to six years of Spring offerings in Purdue's CS2, the second course in the university's introductory programming sequence, Argus found that 45% of students enrolled in Spring 2026 showed patterns consistent with AI-assisted coding. The researchers also compared flagged students against their scores on in-person, proctored written exams.
The comparison is the real finding: students flagged for AI use scored significantly worse on exams where no AI was available, suggesting outsourced homework correlates with weaker actual skill. That's the quiet crisis every CS department now faces, with grades that look fine but understanding that doesn't exist. The researchers are explicit that Argus alone isn't a verdict; they call for mandatory human review before any academic integrity case goes forward, because false accusations carry real consequences for both students and instructors.
One detection tool catching nearly half a class is less a scandal than a baseline reading of how differently CS2 is taught now.