The Real Lesson From Brown's AI Cheating Story Isn't About Honesty
A Brown professor gave his class a take-home exam. The average came back a suspicious 96 out of 100. Concerned, he assigned the same students an in-person final on identical material, with no AI access allowed. The average dropped to 48.
That swing is the clearest piece of evidence to surface this year that AI-assisted take-home work and real learning are not the same thing, and it has kept this story generating coverage for three weeks. El País and Fortune broke it in late June. Breitbart, Yahoo, and Futurism picked up the resolution in early July. Inside Higher Ed reported on July 8 that a survey of 105 Brown faculty found three-quarters concerned about AI cheating on their own campus.
Last week, the story moved somewhere new.
Where the story went next
On July 11, the Boston Globe published an opinion piece extending the Brown case beyond one campus, arguing that schools in general, not just Brown, have unclear rules about where AI assistance ends and academic dishonesty begins. It is the first major outlet to pull this specifically into a K-12-inclusive frame rather than treating it as a higher-ed story.
On July 7, Fortune ran a companion piece with a different angle: the real failure underneath the scandal headlines is not the students. It is assessment design built for a pre-AI world, work that can reward the appearance of mastery without requiring the mastery itself.
Both pieces partially come to the right conclusion. The Globe widens the frame but does not offer a framework. Fortune correctly names design as the problem but writes for administrators setting policy, not for a teacher planning next week's assignment.
What the number actually proves
The 96-to-48 swing does not prove these students are unusually dishonest. It proves that a take-home format, once a capable AI tool exists, stops measuring learning and starts measuring access to that tool. The 48 is not the scandal. It is the true baseline the 96 had been quietly hiding.
That distinction changes what the right response is. Detection software and honor-code enforcement chase a tool that keeps getting more capable every semester; a race that cannot be won on those terms. Redesign changes what the assignment is asking for. A task that requires visible reasoning, a live defense of a position, or application to a scenario built after the assignment is assigned cannot be outsourced the quiet way a take-home essay can, because the thinking has to happen somewhere a teacher can actually see it.
A concrete next step
Every classroom has some version of the Brown take-home sitting in a lesson plan right now. Sorting which assignments are fine as they are, which need a small adjustment, and which need a full redesign is not guesswork. It is a Green, Yellow, and Red triage question, worth running on an assignment before it goes out rather than after a suspicious average shows up.
The Assignment Triage Template, part of the free CALM AI Framework PDF, walks through exactly that process: download it here.
For teachers who want to think through what this looks like in their own building alongside other educators doing the same work, the AI Faculty Break Room is a free community built for that conversation.
I hope to see you there!
Sources:
Primary: Boston Globe opinion, July 11 and Fortune, July 7 ·
Original story: El País, June 28 and Fortune, June 29 ·
Resolution reporting: Inside Higher Ed, July 8