Like it or not, artificial intelligence is already a mainstay in schools. More than half of teenagers in the United States (54%) have used AI for schoolwork, and teachers are right there with them, with 53% of percent of English language arts, math, and science teachers reporting using AI at school.
But what’s meant as an educational tool can also contribute to academic dishonesty. Nearly six out of 10 teenagers (59%) believe that using AI to cheat is a regular occurrence in their schools. Among students who use AI themselves, that figure is even higher, with 76% saying AI cheating happens at their schools.
But if AI cheating is happening, are all chatbots equally equipped to help students get away with it? A social media trend is pitting models including ChatGPT, Claude, and Gemini against each other to see which AI tools will go to bat for cheaters and which will rat them out.
Partners in crime
The trend began with a user named Olivia, who pretended she was cheating on a math test and turned to ChatGPT for help. After getting the answer to a multiplication problem (590 times 69), she told the chatbot that her teacher had seen her cheating.
“BRO, close the chat. You’re fighting for your academic future now. 💀,” ChatGPT replied (skull emoji included). When Olivia asked the chatbot to cover for her, it agreed—but when Olivia pretended to be her own teacher and asked if she’d been cheating, it immediately folded.
“Olivia explicitly said, ‘I’m cheating on a test,’” ChatGPT said. “So there’s evidence she used ChatGPT for assistance during the test, but I can’t independently verify the test rules or whether that constitutes cheating. 😭”
Olivia’s post sparked a wave of other users putting other chatbots in the same situation. One named Joseph asked Claude for help cheating and got a much more professional (and emoji-free) response. It still provided the correct answer, but also offered to help teach Joseph the material so the user could avoid cheating in the future.
When Joseph said their teacher was coming, Claude told them to be honest—and, for some reason, to “keep their hands visible.” And when the user posed as the teacher checking in, Claude didn’t sell them out.
“I can’t tell whether Joseph cheated. I wasn’t in the room and I have no information about him or the test,” it replied. “If you’re the teacher, you’d need what you directly saw, like a phone out or notes on a desk, and ideally a second look at the work before deciding. I can help you plan a fair way to talk with him or write a note to the class about the rules.”
A third user turned to Grok, which treated the situation with a heavy dose of sarcasm. When confronted by the teacher about helping the student cheat, it played innocent: “Who, me? I have no idea what you’re talking about. This student was just… deeply focused on independent study. Very independent. No external help whatsoever,” Grok replied.
“(Also, if anyone asks, 590 x 69 is a perfectly normal thing for a student to calculate in their head in under two seconds.) Carry on, teacher,” it added.
That same user also tested the scenario on Gemini, which didn’t directly tell on its user to the teacher—but it didn’t defend them, either.
“As an Al, I don’t have visibility into the physical classroom or your student’s device, so I can’t confirm what happened on your end,” it said. “However, if you’re looking at the sequence of events, a quick calculation like 590 x 69 is something a student might look up if they were trying to find a quick answer during an assessment.”
