How did the invisible prompt work?
The trap worked by placing an instruction in white text inside a midterm question, making it invisible to students while remaining part of the text copied into a chatbot.
According to a TechSpot report on Gibson’s TikTok posts, the visible question concerned the Industrial Revolution. The concealed instruction told the AI to include information about Madagascar in a way that made no sense.
Gibson, a history professor at Alcorn State University, says 32 of 35 students across two classes copied the question into a chatbot and then submitted its output. He said all 32 failed that portion of the midterm.
In a follow-up video, Gibson showed examples that shifted abruptly from relevant historical discussion to phrases including “Madagascar floats sideways through the afternoon” and “Madagascar purple bicycle whispers to the ceiling.” The examples were presented through Gibson’s posts and have not been independently verified against the graded submissions.
Gibson said he confronted the students and allowed them to contest their grades. According to his account, two chose to do so.
What does the incident establish about AI cheating?
The incident demonstrates how a planted instruction can identify students who copy chatbot output without checking whether the answer is coherent or relevant.
Gibson’s method appears to have detected one specific behaviour: copying an entire prompt into a chatbot and submitting the resulting text with little or no review. It cannot establish how often students use AI more selectively, whether every flagged submission involved the same process, or how widespread the behaviour is beyond these two classes.
The account also rests primarily on Gibson’s own TikTok videos. TechSpot did not cite a statement from Alcorn State University, and the exact course and date of the midterm were not provided. The central numbers should therefore remain attributed to Gibson rather than treated as independently confirmed institutional findings.
Even with those limits, the underlying failure is clear. Students who submitted sentences about Madagascar in an Industrial Revolution response apparently overlooked obvious signs that the generated material was unsuitable. The trap exposed weak verification as much as unauthorised AI use.
What should students and educators take from the case?
Students should verify every AI-assisted answer against the assignment, course material, and applicable academic-integrity rules before submitting it.
A chatbot can produce polished language while following instructions that the user did not notice or understand. Reviewing an answer solely for grammar is therefore insufficient; names, claims, relevance, and reasoning all require scrutiny. Where an institution or instructor prohibits AI assistance, editing the output does not resolve the underlying academic-integrity issue.
For educators, the white-text technique offers a narrow detection method rather than a complete assessment strategy. It may catch direct copying, but students can defeat it by inspecting the prompt or reading the generated answer. Its longer-term value lies in illustrating why assessment rules must specify permitted AI use and why students may need to explain how they reached an answer.
Gibson’s account has travelled widely because the planted references were conspicuous. The more durable lesson is less theatrical: an AI-generated response remains the submitter’s responsibility, and fluent wording cannot substitute for subject knowledge or careful review.
Leave a Reply