Brown Professor Alleges AI-Assisted Mass Cheating in Exam
Illustrates how generative AI can invalidate assessment assumptions - relevant to agencies designing AI capability training or workforce certifications.
Key points
- A Brown University professor alleges mass AI-assisted cheating after 40 of 86 students scored 100 on a take-home exam.
- In-person re-examination produced an average of ~48%, suggesting take-home scores measured prompt skill rather than independent reasoning.
- Limited direct relevance to APS operations; more pertinent to training and certification design than federal AI governance.
Implications for Australian agencies
- Consider Agencies designing AI literacy assessments, workforce certifications, or training evaluations may want to consider whether take-home or unproctored formats remain fit for purpose given LLM access.
Implications are AI-generated. Starting points, not advice — see methodology for how they're framed.
View original source
Copied.
"Brown Professor Alleges AI-Assisted Mass Cheating in Exam"
Source: Let's Data Science – AI Governance
Published: 9 July 2026
URL: https://letsdatascience.com/news/brown-professor-alleges-ai-assisted-mass-cheating-in-exam-e33f4114
A Brown University economics professor alleged in 2026 that widespread AI use distorted a take-home midterm, where 40 of 86 students scored 100 and the class average reached 96%. After switching to an in-person final, the average fell to approximately 48%. The case highlights a structural evaluation problem: tasks solvable by prompting a general-purpose model measure tool access and prompt skill rather than independent reasoning. For practitioners, the lesson is that high-stakes assessments now require explicit policies around tool access, authorship evidence, and process verification - whether in academic or professional contexts.
Implications for Australian agencies:
- [Consider] Agencies designing AI literacy assessments, workforce certifications, or training evaluations may want to consider whether take-home or unproctored formats remain fit for purpose given LLM access.
Retrieved from SIMS, 18 July 2026.