An AI-supervised remote exam went so badly that 58,000 students must retake it
UNAM administered its entrance exam remotely for the first time using LockDown Browser and AI webcam proctoring, resulting in a massive cheating scandal Top scores surged dramatically: 16.3% scored 100+ (vs. 3.5% historically) and 5.5% scored 110+ (vs. 0.9% historically) An expert commission recommended a mandatory in-person "control exam" for approximately 58,000 affected applicants AI proctoring and lockdown browser measures failed to prevent widespread cheating, with experts estimating nearly
Analysis
TL;DR
- UNAM administered its entrance exam remotely for the first time using LockDown Browser and AI webcam proctoring, resulting in a massive cheating scandal
- Top scores surged dramatically: 16.3% scored 100+ (vs. 3.5% historically) and 5.5% scored 110+ (vs. 0.9% historically)
- An expert commission recommended a mandatory in-person "control exam" for approximately 58,000 affected applicants
- AI proctoring and lockdown browser measures failed to prevent widespread cheating, with experts estimating nearly half of students may have cheated
- Cheating tips circulated online advising students to position AI devices outside camera range, hide earphones, or hire substitutes
Why It Matters
This case demonstrates the critical limitations of current AI-powered remote proctoring systems in high-stakes testing environments, raising serious questions about the reliability of automated surveillance for academic integrity. The UNAM scandal serves as a cautionary tale for educational institutions worldwide considering remote examination models, highlighting that technological safeguards alone cannot guarantee fairness when determined actors find workarounds.
Technical Details
- UNAM used LockDown Browser from Respondus, which prevents printing, copying, web navigation, application access, and minimization during exams
- Territorium's AI-powered webcam proctoring system monitored for proxy test-takers, cell phone/earphone use, and applicants leaving the frame
- Human supervision ratio was one supervisor per 150 applicants, with AI-generated alerts for irregularities
- The exam was multiple-choice format, making traditional AI-cheating detection (e.g., paste patterns) ineffective
- Statistical modeling by AI expert Raul Rojas suggested approximately 50% of online test-takers were cheating
Industry Insight
- Institutions should treat AI proctoring as a supplementary layer rather than a standalone solution for remote exam integrity
- The UNAM case validates concerns about "arms race" dynamics between cheating methods and surveillance technology, suggesting hybrid in-person/remote models may be more reliable for high-stakes assessments
- Educational organizations should invest in behavioral analytics and longitudinal assessment data rather than relying solely on real-time monitoring tools that can be circumvented
Disclaimer: The above content is generated by AI and is for reference only.