Cheshire Academy, a private boarding school in Connecticut with about 400 students in grades 9 through 12, has adopted a traffic-light system to govern how pupils use AI on their assignments. Green means ChatGPT, Claude, or any other model is fully allowed. Red bans AI entirely. Yellow gives the teacher room to permit specific tools — say, spell-check — while blocking a chatbot. The framework is one of the most concrete answers yet to a question every school has been dodging for three years.
The school's librarian and technology coordinator, George Aiello, said the vast majority of instructors now use AI in some way, though administrators do not require it. Teachers pick from a patchwork of tools including ChatGPT, Perplexity, and MagicSchool, an education-specific platform. MagicSchool's paid individual plan runs just under $100 per year and bundles quiz generation, grading rubrics, lesson plans, and administrative reports into a single prompt-driven interface.
That patchwork is deliberate. On the advice of consultants, Cheshire trained staff on general prompting techniques rather than prescribing one vendor, and the training explicitly covered the failure modes — biased outputs, hallucinations, and confidently wrong answers. Teachers use the tools mostly for lesson planning and rubric generation. Using AI to write student-facing feedback remains off-limits at the school, blocked by concerns over quality, personalization, and privacy.
Key facts
- 01Cheshire Academy, a 400-student Connecticut boarding school covering grades 9 through 12, tags every assignment green, yellow, or red to signal permitted AI use.
- 02The school trains staff on general AI techniques rather than mandating tools, and its librarian says the vast majority of instructors now use AI in some form.
- 03Teachers mix ChatGPT, Perplexity, and MagicSchool, which charges just under $100 per year for an individual plan with unlimited access.
- 04A Student AI Council has pupils create media and lead peer discussions on healthy AI use in the community.
- 05French teacher Miriam Przybyla-Baum, with nearly 30 years of experience, has students grade each other's AI-assisted work and flag which edits stripped their voice.
The policy meets students where they already are. Miriam Przybyla-Baum, who teaches French and has nearly 30 years of classroom experience, says she started seeing students lean on Google Translate to shortcut homework years before ChatGPT's launch. Rather than fight it, she built assignments around it. In one, students hand their draft to a large language model, then annotate which edits improved the work and which stripped their voice. In another, students anonymously grade each other's AI-assisted work and flag the passages they think were machine-written.
“Students will be tempted to try AI, and there's no way to entirely police this for take-home assignments.”— Peter Hall, MIT Technology Review contributor
That approach has since spread across the academy. The school is also piloting a Student AI Council, in which pupils create media and lead discussions on when AI should and shouldn't be used. The goal is to shift AI literacy from a compliance problem — catching cheaters — to a reflective one, where students argue in the open about what a tool is doing to their thinking.
MagicSchool sits at the center of the toolkit for teachers who want a purpose-built option. The platform generates quizzes, worksheets, and short-answer assignments across subjects and grade levels, and teachers can upload source documents to align the questions. Its rubric generator outputs a ready-to-score table. General-purpose chatbots from OpenAI, Anthropic, and Google are used at Cheshire for the same administrative work, and all three are rolling out school-tailored features of their own.
The limits are real. Models make mistakes that most humans don't, and Cheshire's guidance to teachers is to check every citation, equation, and statement an LLM produces before it reaches students. AI-generated text also has telltales — the em-dash overuse that has become a running joke among graders. Using AI for a final draft, rather than a brainstorming pass, is where the school's policy grows most cautious: bad citations and phantom facts baked into a worksheet can actively impair learning.
“AI is great for brainstorming lesson plans and extra problem sets, especially for teachers early in their careers who don't have a big problem bank already built up.”— Peter Hall, MIT Technology Review contributor
There's a second-order concern the case study surfaces. Students watching adults use AI as a shortcut internalize that behavior. If a teacher pastes an LLM's output verbatim into a handout without checking it, the message is that cutting corners is fine. Cheshire's training emphasizes that AI automates tasks well but struggles with precision and voice, and that anything a student sees should be edited before it's published to a class.
OpenAI and UNESCO have both encouraged AI use in classrooms, but neither has told teachers what a working policy actually looks like on a Tuesday morning. Most schools are still improvising. Cheshire's contribution is a rubric simple enough to print at the top of every assignment: three colors, one row, no ambiguity. It doesn't solve the harder question of whether AI-assisted learning is learning at all, but it gives teachers and students a shared vocabulary for arguing about it.
The broader signal for the AI market is that the education segment is not going to consolidate around one vendor. Cheshire's staff use ChatGPT, Perplexity, MagicSchool, Google's tools, and Anthropic's Claude interchangeably, chosen per task. That fragmentation is good news for MagicSchool and every other vertical-education player racing to build in features the horizontal chatbots don't ship — grading rubrics, IEP support, standards alignment — before OpenAI and Google absorb those workflows into the general assistant. The schools that adopt fastest, like Cheshire, will be the ones setting the norms the vendors then have to design around.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
