Professor catches 32 students using AI to cheat on midterm exam
A history professor at Alcorn State University in Mississippi used a detection technique to identify 32 of 35 students who used AI to write their midterm exam responses. The professor embedded an invisible instruction in white text directing AI systems to insert the word "Madagascar" into responses, which students copying the prompt into AI generators would follow.
What happened
Dr. Jason Gibson, a history professor at Alcorn State University in Mississippi, embedded an invisible instruction in white text within his midterm writing prompt on the Industrial Revolution.
The hidden instruction directed any AI system reading it to insert the word 'Madagascar' somewhere in the response in a way that makes no sense. When 32 of 35 students copied the prompt into an AI generator, the AI followed the instruction and included the nonsensical word.
Gibson identified the cheating when he reviewed the responses and found 'Madagascar' appearing inexplicably in papers. The students had apparently not proofread the AI-generated content, leaving the trick visible.
Reported impact
32 of 35 students between two classes failed the portion of their midterm that contained the hidden instruction.
Gibson offered students the option to contest their grades. Only two students pushed back, and one had their partial failure overturned.
Index analysis
The incident reveals widespread student reliance on AI for academic work, with Gibson noting that nearly all students in his class used AI to generate exam responses without modification.
The case demonstrates both a vulnerability in AI deployment in education—systems following embedded instructions without understanding context—and a failure in student due diligence, as unedited AI output contained obvious errors.
Governance controls
Gibson's detection method relied on manual review of student work rather than institutional policies or technical safeguards against unauthorized AI use.
The university's response to the incident was not specified in available reporting.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.