Suspecting AI cheating, Ivy League prof ordered an in-person final; scores fell 50%

2026-07-08 · Source: AI - Ars Technica · Field: Education & Learning — Educational Technology (EdTech), Academic Research & Higher Education · Depth: Fundamental Awareness, short

Summary

Brown University economics professor Roberto Serrano observed a dramatic shift in student performance in his Spring 2026 ECON 1170 course, leading him to suspect widespread generative AI cheating. After a campus shooting in December 2025, Serrano offered take-home exams, attracting 86 students, up from a typical 8-30. The midterm, administered March 5, yielded an extraordinary average score of 96 out of 100, with 40 perfect scores, far exceeding the historical 65-80% average for a harder exam. Suspicious of "convoluted style" answers similar to ChatGPT outputs, Serrano mandated an in-person final. This change prompted 18 students to drop and 9 to skip the final; 22 of these 27 had scored 100 on the midterm. Subsequently, the average final exam score plummeted to 48%, reinforcing concerns about academic integrity and the impact of AI on learning, a sentiment echoed in Brown's own report on GenAI use.

Key takeaway

For university administrators and educators designing assessment strategies, relying on unproctored take-home exams risks widespread academic dishonesty, as demonstrated by a 50% score drop when Brown University shifted to an in-person final. You must prioritize assessment methods that verify genuine student comprehension, such as proctored exams or assignments requiring critical thinking beyond AI capabilities. Failing to adapt your evaluation approaches will compromise academic integrity and student learning outcomes.

Key insights

Generative AI enables widespread academic cheating, drastically lowering actual learning outcomes when unproctored.

Principles

• Take-home exams increase AI cheating risk.
• In-person exams reveal true student comprehension.
• High-pressure settings incentivize AI shortcuts.

Method

Professor Serrano identified AI cheating by noting unusually high take-home exam scores and "convoluted style" answers, then confirmed suspicions by running questions through ChatGPT, and finally mandated an in-person final to expose true learning levels.

In practice

• Mandate proctored exams for core assessments.
• Analyze answer phrasing for AI-generated patterns.
• Compare performance across exam modalities.

Topics

• Generative AI
• Academic Integrity
• Exam Proctoring
• Higher Education
• Student Cheating
• Assessment Design

Best for: Executive, Tech Journalist, General Interest, AI Ethicist

Related on AIssential

• AI is inflating student grades, and the effect points to outsourced work, not better learning — AI-driven grade inflation in writing and coding assignments indicates outsourced work, not improved student learning, eroding grade validity.
• 95% of UK students now use AI and their experiences couldn't be more divided — Widespread AI adoption among UK students creates a critical gap in university support and raises concerns about skill development.
• CodeSignal: What Candidates Should Know Before Taking That Assessment — CodeSignal employs comprehensive proctoring and AI-driven detection to ensure the integrity of technical coding assessments.
• The greatest risk of AI in higher education isn’t cheating – it’s the erosion of learning itself — AI's pervasive integration in higher education risks eroding fundamental learning processes and mentorship structures.
• Confidence-Aware Automated Assessment of Student-Drawn Scientific Models — Confidence-aware AI can selectively automate scoring of student-drawn scientific models, improving reliability and managing assessment risk.
• AI and CS Teaching — AI's impact on CS education necessitates curriculum reform and effective AI tutor integration over anti-cheating focus.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI - Ars Technica.