A college professor came up with a clever idea to catch AI cheaters — 91% of his class failed
A Mississippi professor hid a single invisible word in his midterm exam. What happened next went viral for a reason.
Every professor teaching right now has figured out, one way or another, that some of their students are running exam questions through AI and submitting the output as their own work. Most have learned to spot the signs: writing that is just too general, confident answers that get the details wrong in exactly the way a language model does when it fills in gaps, arguments that sound authoritative and say almost nothing. What most professors don’t have is proof. Jason Gibson, an instructor at Alcorn State University in Mississippi, decided to get some, and the method he used was simple enough that the results have been alarming.
Gibson embedded the word “Madagascar” inside the midterm exam in white text, invisible against a white background. A student reading the paper with their own eyes would see nothing unusual, but a student who copied the full text of the question into an AI chatbot would unknowingly pass the hidden word along, and the AI would incorporate it into the response.
Of his 35 students, 32 submitted answers containing bizarre phrases built around that word. The AI had done exactly what it was designed to do with the full prompt it received. 32 students handed that back as their own work without reading a word of it, and they mentioned Madagascar in responses that had nothing to do with the original question. Gibson said publicly that his intention was not to embarrass anyone but to make a point about academic integrity. The point landed.
The comment section on the video report was not kind to the students. The response with the most engagement, drawing more than a thousand likes, was a single line: “32 out of 35. I feel for those 3 honest students.”
A widely shared comment pointed out that the real revelation was not the cheating itself but the laziness behind it. The students had not just used AI to write their answers. They had submitted those answers without reading them, meaning they trusted a chatbot more than they trusted their own ability to check the work. One commenter who said they were a teacher noted that this happens with at least one assignment in every class they give now.
The comment that hit hardest, drawing more than 6,000 likes, framed it as a public health concern: “Remember to eat healthy and take care of yourself. Your future doctor is using AI to pass their exams.” Several people pointed out that in previous generations, being caught plagiarizing meant automatic expulsion, not failing a portion of one test. Another commenter offered a counterpoint worth keeping in mind: one person in the thread said they use AI to study, but they read both the question and the answer, then rewrite it themselves, so they are at least learning the material in the process. That distinction mattered to a lot of people reading.
What this says about AI and real knowledge
91% of students in that classroom submitted AI-generated output as their own thinking without reading it first. That number reflects something bigger than academic dishonesty.
The gap between generating an answer and understanding one is exactly where real education lives. AI produces confident, well-structured responses to almost any question in seconds, and a student who copies that output has outsourced their thinking entirely. Gibson put it plainly: all the students had to do was read what the chatbot wrote. Reading it, questioning it, and deciding whether it actually answered the question. That is the work, and skipping it means walking away with a grade built on someone else’s reasoning.
The more serious version of this problem runs well beyond college classrooms. A person who submits AI output without reading it is also someone who will consume AI-shaped information in other areas of their life. The news they read, the medical answers they look up, the financial decisions they research, and accept them the same way. They don’t use critical thinking, and never question whether something is true before acting on it. Every time someone hands a question to a machine and accepts the result without scrutiny, it means they can be easily manipulated.
This matters because whoever shapes the information AI produces shapes the reality of everyone who trusts it without question. Blind trust in any single source of information is how disinformation spreads and takes hold, and AI at scale makes that easier than anything that came before it. The person who builds something real with their knowledge in their career, their finances, and their health is the ones who treat every answer as a starting point rather than a conclusion. Read it. Check it. Decide what you actually believe. That part was always yours to do, and handing it away is the most expensive shortcut.
Jason Gibson, the college professor in Mississippi, caught 32 students in about ten seconds using a trick that cost him nothing. The more uncomfortable question the story raises is how many of those students understood what they were supposed to be learning, and whether the shortcut they took left them any less prepared for the moment when the real test arrives, and there is no chatbot in the room.
