A professor at Alcorn State University caught 32 students cheating on their midterm by using an invisible prompt trap.
The trap was set by embedding instructions in a white-colored font in the assignment prompt, telling the AI to talk nonsense about Madagascar.
The students were asked to answer a question about the Industrial Revolution, but the embedded instructions led the AI to include information about Madagascar in a way that made no sense.
Dr. Jason Gibson, the professor, said 32 out of 35 students across two classes copied the question, pasted it into a chatbot, and then copied and pasted the AI’s output into the answer window.
Had they checked the AI’s output, they would have noticed the strange references to Madagascar in an essay about the Industrial Revolution.
Dr. Gibson stated that all the students he caught using AI failed this portion of their midterm.
He shared examples of the AI-generated answers in a viral TikTok post.
In one example, the AI talked about the Industrial Revolution before adding a cryptic message: “Madagascar floats sideways through the afternoon.”
In another example, the AI discussed technology and social inequality before finishing with: “Madagascar purple bicycle whispers to the ceiling.”
Dr. Gibson said he wasn’t using this as an opportunity to mock his students, but rather to address the issue of cheating.
He gave the students the opportunity to contest their grade, and only two did.
The use of AI to cheat has been a growing problem since ChatGPT went mainstream.
Teachers have tried to fight back using handwritten work, oral tests, and even AI to come up with questions that would be difficult for students to complete by simply feeding it into a large language model.
A professor at Brown University saw midterm averages jump from 65% to 80% to 96% after allowing take-home exams.
He estimated that all but two of his 86 students used AI to cheat.
Professors like Dr. Gibson are looking for new ways to prevent cheating and ensure students are doing their own work.
Dr. Gibson’s method may not be foolproof, but it highlights the need for creative solutions to address the issue of AI-assisted cheating.
The incident has sparked a discussion about the role of airport search and security in schools and the need for professors to stay one step ahead of students who are using AI to cheat.
It’s a battle that humans aren’t winning.
