Overview
- Dr. Jason Gibson at Alcorn State said he hid the word “Madagascar” in white text inside a midterm prompt and that 32 of 35 students returned answers containing the nonsensical Madagascar phrases after pasting the question into chatbots.
- Gibson told students they could appeal their grades and said two did so, with one grade changed after the student said dark mode had kept the hidden text from view.
- A separate TikTok teacher using a hidden “Gatorade” prompt reported that an AI detector flagged a paper as about 70% AI-generated, indicating similar copying of prompts into chatbots without proofreading.
- Teachers and professors are shifting test formats toward handwritten or in-person work and requiring AI disclosure while warning that detection tools give probabilistic results and that models can evolve to resist simple prompt injections.
- The incidents show direct consequences for students who outsource answers and signal likely policy changes on assessment, grading appeals, proctoring, and how schools teach responsible AI use.