An Ivy League professor reported that a large share of students produced near-perfect scores after switching to take-home exams, then used ChatGPT internally to evaluate whether AI-assisted work was shaping student submissions. The case highlights how generative AI can undermine standardized measurement of learning outcomes even when institutions add safeguards. The professor’s findings—students producing scores that surged and exam solutions that appeared convoluted when compared to model-generated reasoning—prompted a move away from take-home exams as reliable assessment evidence. The report positions assessment integrity as a key academic governance concern as AI-enabled cheating scales. Beyond cheating detection, the article underscores a more structural assessment issue: institutions that adopt open-ended AI-friendly learning practices may also need to redesign assessment models, including question formats, proctoring alternatives, oral defenses, and evaluation rubrics that measure individual reasoning. For higher education leaders, the immediate impact is operational: faculty and assessment committees must treat AI as a measurement problem, not only an academic integrity policy issue.