A student submits an essay. Well-structured, clearly argued, grammatically flawless. It gets an A. But the student didn't write it. They didn't even think it. They prompted it, cleaned it up, and hit submit. The grade reflects prompting skill, not learning.
Two major studies now put real numbers behind this problem.
The Berkeley Study: 30% More As
In May 2026, Igor Chirikov at UC Berkeley's Center for Studies in Higher Education published "Artificial Intelligence and Grade Inflation". He analyzed over 500,000 enrollments across 319 courses at a large research university between 2018 and 2025. The finding: courses exposed to generative AI saw the share of A grades rise by 13 percentage points, a 30% increase from the pre-ChatGPT baseline.
The effect hit hardest in writing-heavy humanities courses and coding-heavy CS courses, basically anywhere take-home work carried significant grading weight.
Chirikov's conclusion: "The results suggest that students have relied on generative AI to do better in their studies, not that these classes are learning more."
Grades went up. Learning did not.
Harvard: When 60% of Grades Are As
At Harvard, over 60% of all grades are now As, up 25% from two decades ago. In early 2026, Harvard faculty voted on a proposal to cap A grades at 20% per course.
The problem is that capping grades doesn't actually fix anything. A Harvard Crimson analysis made this point clearly: when students can produce A-quality text without A-quality thinking, you're just redistributing letters. The disconnect between grade and learning stays the same.
The faculty knows the cap is a stopgap. How do you assess learning when AI can produce the output without the thinking? Nobody has a great answer yet.
The Real Problem: Output Doesn't Equal Learning Anymore
Traditional grading assesses the quality of submitted work. That used to be fine, because output quality and learning were correlated. AI broke that correlation.
A polished essay used to mean a student who read the material, formed an argument, struggled with structure, and revised for clarity. Now it might just mean a student who typed a good prompt.
Same grade. Completely different learning.
What This Means for Teachers
If you're a teacher, you already know what's happening. You've seen the suspiciously perfect essays. You've noticed students who can't discuss their own papers in class. You've wondered whether grades mean anything anymore.
The research confirms what you're seeing. And it points in a specific direction: you need to see the process, not just the product.
When you can see how an essay was actually written, the thinking pauses and revisions and struggles with that third paragraph, the grade means something again. A student who spent four hours wrestling with an argument and produced a B paper learned more than one who prompted a perfect essay in ten minutes.
What Limudi Does About This
We built Limudi around this idea. We don't look at a finished submission and try to guess what happened. We watch the writing as it happens.
When a student writes in Limudi, you see the full journey from blank page to final draft. The AI Judge gives you a verdict on authenticity and explains its reasoning, based on the actual behavioral record of how the work was composed.
Students can still use AI. We built an AI tutor right into the platform. But the tutor guides their thinking, asks them questions, and pushes them to develop their own arguments. It doesn't write for them.
The grade reflects what the student actually did. That's what we're solving for.
Where This Leaves Us
AI broke the grading system's core assumption: that output quality reflects learning. Berkeley put numbers on the damage. Harvard is trying to cap grades but knows it's not enough.
Banning AI won't work. More detectors won't work. Making the writing process visible again will. That's what lets grades mean what they're supposed to mean.
References:
- Chirikov, I. (2026). "Artificial Intelligence and Grade Inflation." UC Berkeley Center for Studies in Higher Education. escholarship.org/uc/item/80x8d3qd
- Harvard Faculty of Arts and Sciences (2026). Grade inflation proposal. Coverage: Inside Higher Ed, Fast Company
- Previn, S. (2026). "How AI Undermines Harvard's Grading Reform." The Harvard Crimson