Mississippi professor comes up with genius way to catch his students out for cheating with AI
A Mississippi professor has come up with a genius way to catch his students out for cheating with AI, and it meant a whole lot of students failed their midterms.
Dr. Jason Gibson, a history professor at Alcorn State University, secretly planted an invisible instruction inside a midterm exam.
The trick was so effective that 32 of his 35 students ended up giving themselves away.
And the funniest part is that they apparently didn’t even bother reading what the chatbot had written before handing it in.
How this professor hid a sneaky AI trap in his exam
Artificial intelligence use has been increasing in popularity over the last few years, and it’s gone from the idea of futuristic robot butlers to becoming a way that university students get out of writing essays.
A history professor at Alcorn University, Dr. Jason Gibson, became suspicious after noticing that his students’ written work seemed strangely similar, despite their individual writing styles.
So, for a midterm assignment about the Industrial Revolution and the modern digital age, he decided to try something different.
Hidden inside the online exam prompt was text written in white, meaning it blended into the background and was effectively invisible to students reading the question normally.
But if someone copied the entire prompt and pasted it into an AI chatbot, the chatbot could still read the hidden instruction.
That instruction told the chatbot to include the word ‘Madagascar’ somewhere in the response in a way that made absolutely no sense.
The idea was genius: students who completed the assignment themselves would never know about the hidden word, while anyone blindly feeding the question into a chatbot could end up with a very strange reference to the African island nation, and Gibson’s trap worked remarkably well.
When he started marking the responses, he found bizarre references to Madagascar appearing throughout answers that were supposed to be about history just due to the hidden chatbot prompt.
One response reportedly included the bizarre line ‘Madagascar floats sideways through the afternoon’, while other students managed to work the word into similarly nonsensical sentences.
The references had nothing to do with the subject matter, making them a pretty obvious giveaway.
What happened when the students were caught out
Gibson says 32 of the 35 students involved across his classes included the hidden word in their answers, meaning more than 90 percent fell into the trap.
Those students failed that portion of the midterm, according to Gibson, although he emphasized that the point that his genius idea wasn’t to embarrass anyone.
Instead, he wanted to demonstrate the importance of academic integrity and show students just how easily blindly relying on AI can backfire.
The really remarkable part, however, was that the students apparently didn’t spot the nonsensical sentences before submitting their work.
That means Gibson’s trick didn’t just reveal who had used A, it also showed how many students were apparently willing to submit AI-generated answers without checking whether they actually made sense.
The professor’s approach has since gone viral, highlighting the increasingly complicated battle between educators and students as AI becomes more common in schools and universities.
AI detection software can be unreliable, while students can use tools designed to make AI-generated writing appear more human.
Gibson therefore found a remarkably low-tech but genius alternative: give the chatbot an instruction that a genuine student would never see.
The method may sound almost too simple, but the results speak for themselves.
Of course, AI can be a useful tool for research, brainstorming, and learning when used appropriately, but if you’re going to hand in an AI-generated answer, at least reading it first might be a good place to start.