Overview
- Alcorn State history professor Dr. Jason Gibson posted on TikTok Monday showing he hid white-font instructions referencing “Madagascar” in a midterm prompt and said 32 of 35 students failed part of the exam after chatbots reproduced the hidden text in their answers.
- The trap worked because many students copied the full question into a large language model and pasted the AI output back without proofreading, which caused nonsensical Madagascar phrases to appear and reveal AI use.
- A recent Higher Education Policy Institute report found automated AI-detection tools are often inaccurate and prone to false positives, prompting universities such as UCLA and Waterloo to stop using some popular detectors.
- Some schools are reverting to offline, supervised tests and device bans while others are redesigning assessments with staged submissions, oral defenses and embedded AI literacy to make student learning visible and verifiable.
- Gibson offered students an appeal, two contested their grades and one was changed after a student said dark mode hid the white text, an outcome that highlights fairness concerns and is likely to accelerate changes in grading policy and classroom practice.