CS240 AI Cheating Retrospective

Professor Jeff Turkstra recounts a large academic-integrity episode from CS 240 (Spring 2026) centered on student use of AI/LLM tools to complete assignments. The course had an explicit ban on AI-generated solutions, repeated in the syllabus and multiple lectures. Detection combined traditional similarity tools (MOSS) with a custom static-analysis detector called Argus, which became usable mid-March; an "AI Academic Integrity" (AIAI) team formed March 23 to manage cases. Rather than immediate punitive action, a self-reporting mechanism modeled on prior practice was used: emails on April 16 required flagged students to complete a response form by April 20 or face an F and notification to the Dean. Of 599 starters (65 withdrawals), roughly 267 of 584 identified students were flagged (~45.7%), with about 207 current enrollees and 60 dropped students contacted.

Argus is clarified as a static-analysis tool, not a custom LLM; a preprint and conference submission exist but the SIGCSE submission was not accepted. Analysis showed the indicators Argus relies on were virtually absent before 2024 and then rose rapidly, producing high H-scores; every flagged case underwent human review and many were not pursued. Turkstra acknowledges shortcomings in how the situation was handled, explains the conservative adjudication approach, and invites vetted educators to request indicator details rather than publicizing them.