GlobeskoolThe AI-Education Death Spiral: When 'Let Them Cheat' Becomes the Plan
Princeton dropped its 133-year honor code over AI cheating and replaced it with proctors. That is not a fix. Here is what the real skill gap looks like, and what parents can do about it.
Princeton ended its 133-year-old honor code last year because of AI-assisted cheating. The solution the school landed on: bring proctors back into every exam room. Stanford moved in a similar direction. Other institutions quietly rewrote honor policies they had not touched in decades.
Nobody asked the obvious question. If you have to physically watch a student think in order to confirm that they are actually thinking, something upstream already failed.
The Problem Is Not the Kids Using AI
Here is the part that tends to get lost in the policy conversation. The death spiral is not students reaching for a tool that produces fast answers. That is a rational response to an environment where the only thing being measured is the answer itself.
For decades, most school assessments were built around a single deliverable: produce the output. Write the essay. Solve the problem set. Hand it in. That design is functional when the only instrument available is the student’s own brain. The moment a better answer-producer shows up, the whole structure collapses.
AI did not break that model. It revealed that the model was already fragile.
Wharton researchers published a working paper using the phrase “cognitive surrender” to describe what is happening at the university level. The phrase is accurate, but it is worth being precise about who is surrendering. It is not primarily the students. It is institutions that spent years testing retrieval and recall, and are now acting surprised that a retrieval-and-recall machine showed up and outperformed their students at the one thing being graded.
What a Proctored Exam Actually Tests
Proctored exams test a specific thing: recall under surveillance. That is not nothing. There are contexts where knowing something cold, without assistance, genuinely matters.
But it is a narrow thing. And it is not the thing that determines what a 14-year-old becomes at 24.
The skills that compound over time are different. Can a student frame a problem before they start solving it? Can they look at an AI-generated answer and identify where the reasoning goes soft? Can they decide when a tool is helping them think and when it is thinking for them? Can they argue with a source rather than just use it?
None of those skills show up on a proctored exam. None of them are particularly cheatable either. You cannot paste a prompt into a chatbox and get back genuine judgment. That is the irony. The skills that are hardest to fake are exactly the ones being least tested.
The Habit Forms Earlier Than Most Parents Realize
The conversation in policy circles tends to focus on higher education because that is where the visible crises are landing. Honor codes abolished. Admissions essays suspect. Dissertations in question.
But the habits form well before college. A kid who reaches for a tool instead of sitting with a hard problem, who copies without reading, who never learns to interrogate output because no adult ever asked them to, arrives at 18 already behind in ways that a proctored exam will not fix.
This is not about banning AI. That ship has sailed and it was never going to work. It is about who is in charge of the tool.
A student who copies AI output and submits it has let the tool do their thinking. A student who reads AI output, annotates it, argues with the parts that seem off, and builds something from that process is doing something categorically different. The physical act might look similar from outside the room. The cognitive experience is not.
What the Fix Actually Looks Like
It starts with changing what gets measured. If the task is “evaluate this AI-generated argument and identify the three places the reasoning is weak,” cheating becomes nearly impossible. You have to know something. You have to think. You have to exercise judgment.
Real projects over passive consumption. A kid who builds something, presents it, defends it, and iterates on it is developing a track record of thinking that no single AI interaction can replicate.
Judgment over recall. Recall is a component of learning, not the destination. The question worth asking about any student is not “can they produce the answer” but “do they know what to do when the answer is wrong.”
The kid in charge of the tool, not the other way around. This is the practical version of AI literacy. Not a unit on how chatbots work. A habit of use where the student is always the one directing, evaluating, and deciding.
These are not supplementary skills. They are the skills. Core subjects build the knowledge base that makes judgment possible. But knowledge without the ability to apply, question, and create with it does not compound the way parents hope it will.
What Parents Can Do Right Now
The school your child attends may or may not be redesigning its approach to any of this. Many are still in the detection-and-lockdown phase. That is not a reason to wait.
The conversation worth having at home is not “are you using AI for homework.” It is “when you use it, what are you doing with the output.” The difference between those two questions is the difference between policing a behavior and building a habit.
Globeskool was built for exactly this gap. Core subjects for kids 8 to 16, plus the Future Skills that standard curricula skip: critical thinking, problem framing, learning how to learn, and working with AI as a tool you direct rather than one that directs you. Real projects. Judgment over recall. The student in charge.
If your child is between 8 and 16, the assessment is worth five minutes of your time.