When chatbots first hit classrooms, the playbook was simple: ban them or ignore them. Neither worked. Cheshire Academy, a private Connecticut boarding school with about 400 students, is now trying a third path—teaching students to think critically about when AI actually helps them learn and when it just does the work for them.
The school never mandated a single AI tool. Instead, staff trained on general prompt-writing techniques and the technology’s failure modes, like its tendency to fabricate facts or bake in bias. Teachers now use ChatGPT, Perplexity, and MagicSchool to plan lessons and build grading rubrics, though none have yet trusted AI with student feedback. Miriam Przybyla-Baum, a French teacher with nearly 30 years of materials behind her, skipped the AI tools entirely and went straight to pedagogy. Her students let an LLM edit their homework, then decide which edits actually improved the work and which stripped out their voice. Another assignment has them anonymously grade each other’s AI-assisted submissions and annotate where they suspect machine involvement.
The school’s signature move is a traffic light labeling system. Green means AI is fully allowed. Red means hands off. Yellow is the interesting one—teachers permit specific tools, like spell-check, while banning chatbots. It’s a deceptively simple framework that forces instructors to articulate exactly what skill they’re testing, rather than issuing blanket AI policies that students ignore anyway.
Cheshire is also piloting a Student AI Council where students create media and lead discussions about healthy AI use. The bet is that peer influence will do what administrative decrees can’t. Whether that works remains an open question, but the school has already solved one problem most institutions haven’t: it stopped pretending there’s a single right answer to AI in education.
