BE HUMAN LABS Talk to us
← From the lab
Point of view

Outsourced thinking is the real risk, not cheating.

Cheating rates did not rise when AI arrived. Something quieter did. Cheating breaks a rule; outsourced thinking skips the part of learning that grows a mind.

Shikhar Anand, FounderOctober 20265 min readOpening only

When ChatGPT was released, schools braced for a wave of cheating. Researchers at Stanford had been surveying American high school students about cheating for over fifteen years, so they were well placed to measure it. Before AI, between 60% and 70% of students admitted to at least one cheating behaviour in the previous month. After AI, the figure stayed about the same, and in some schools fell slightly.

The wave did not come. Students who cheat found a new tool; students who did not, mostly still did not. But that does not mean nothing changed. The real change was harder to see, because no rule was broken.

60-70%

of US high school students reported cheating in the past month, before and after ChatGPT. The rate did not rise.

Challenge Success, Stanford Graduate School of Education
61%

the average rate at which seven popular detectors wrongly flagged essays by non-native English writers as AI-written.

Liang et al., Stanford, 2023
−17%

exam scores for students who openly used AI in practice, with no rule broken.

Bastani et al., PNAS 2025

Talk to us

About: