Glonce

Cheshire Academy sorts AI homework rules with a traffic light system

miscThe Glonce editorsMIT Technology Review🔥53Glonce rating53Impact48Novelty42Relevance78Surprise32Players42Credibility82
Cheshire Academy sorts AI homework rules with a traffic light system

Chatbots caught many schools off guard when they arrived a few years ago: students suddenly carried an app that could answer almost any homework question or draft an essay in seconds. Teachers can often tell anyway, since models make mistakes most humans don't and, according to some teachers, AI generated text has simple tells like overusing em dashes. Still, the generative AI boom added to an already long list of duties for teachers, who were already working long hours planning lessons, writing assignments and grading exams. Organizations from OpenAI to UNESCO encourage AI use in the classroom, but MIT Technology Review reports many teachers remain unsure how to actually apply it.

As a case study, the outlet profiles Cheshire Academy, a private boarding and day school in Connecticut with about 400 students in grades 9 through 12. Administrators there do not require instructors to use AI at all, though the school's librarian and technology coordinator, George Aiello, says a vast majority of instructors use it in some way. Faculty rely on a patchwork of tools: general chatbots like ChatGPT and Perplexity alongside MagicSchool, a platform built specifically for educators. On the advice of outside consultants, the school chose to train staff on general AI techniques, including how to write effective prompts, rather than mandate particular tools, while also stressing the technology's limits and its potential to produce incorrect or biased answers.

Teachers mostly use generative AI to prepare materials such as lesson plans and grading rubrics. Some want to use it to give students feedback too, but concerns about quality, personalization and privacy have kept them from doing so just yet. French teacher Miriam Przybyla-Baum, who has taught for nearly 30 years, does not use AI to build her own materials since she already has a deep library of them, but she works AI directly into her lessons. She started noticing students take shortcuts with tools like Google Translate years before ChatGPT launched, and now has students let a large language model edit their homework, then decide which edits were accurate and which ones erased their own voice. In a separate exercise, students anonymously grade classmates' AI assisted assignments and mark which parts they believe were AI generated.

Cheshire Academy later adopted a version of Przybyla-Baum's reflection exercises school-wide: assignments are now labeled like traffic lights, with green meaning AI use is fully allowed and red banning it outright. Yellow lets a teacher permit some tools while banning others, such as allowing spell-check but not a chatbot. The school is also piloting a Student AI Council, in which students create media and lead discussions about healthy AI use, meant to push students to reflect on how AI should and should not be used to benefit their community.

MagicSchool, which Cheshire Academy previewed to staff as generative AI was going mainstream, packages a wide range of features on one platform. It can generate quizzes, worksheets and assignments across many subjects and grade levels, includes a grading rubric generator that produces a ready-to-use scoring table, and can build presentations, lesson plans and administrative reports. Teachers enter task-specific prompts, such as grade level, question count and question type, and can attach documents to align questions with. The platform has free and paid versions; unlimited access and complete records cost just under $100 a year on an individual plan. Many Cheshire teachers instead rely on general-purpose chatbots from Anthropic, Google, OpenAI and others for administrative tasks, and some of those companies are rolling out features aimed specifically at schools, with mixed results so far.

MIT Technology Review closes with three pieces of guidance drawn from the case study. First, meet students where they are: there is no way to fully police AI use on take-home assignments, so teachers need to counter misinformation about what AI can and cannot do and teach healthy habits instead of relying on bans. Second, model best practices: AI automates tasks well but struggles with precision and voice, and students who see teachers or administrators use it carelessly in anything others will see may take that as license to cut corners themselves. Third, refine and replace: AI is useful for brainstorming lesson plans and extra problem sets, especially for early-career teachers who don't yet have a large bank of material built up, but because it can hallucinate, using its output for final drafts risks confusing students or impairing learning, so every citation, equation and statement still needs to be checked.

Key facts

  • Cheshire Academy, a private school of about 400 students in grades 9 through 12 in Connecticut, labels homework assignments green, yellow or red to set how much AI students may use on each one.
  • The school trained staff on general AI techniques and limits rather than mandating specific tools; librarian and technology coordinator George Aiello says a vast majority of instructors use AI in some way.
  • French teacher Miriam Przybyla-Baum, who has taught for nearly 30 years, has students edit AI-revised homework and anonymously flag which parts of classmates' assignments look AI-assisted.
  • Cheshire Academy is piloting a Student AI Council, where students create media and lead discussions on healthy AI use.
  • MagicSchool, an AI platform for educators used at the school, charges just under $100 a year for an individual plan with unlimited access; many teachers instead use general chatbots from Anthropic, Google and OpenAI for administrative work.

Why it matters

Chatbots blindsided many schools when they arrived a few years ago, and teachers already stretched thin by lesson planning and grading have had to adapt with little clear guidance. Groups from OpenAI to UNESCO encourage classroom AI use, but MIT Technology Review says many teachers still feel confused about how to apply it in practice. Cheshire Academy's structured response, staff training instead of a tool mandate, a simple traffic light labeling scheme, and a student-led council, gives other schools a concrete model to weigh instead of choosing only between an outright ban and unrestricted use.

Who it affects

Directly: the roughly 400 students in grades 9 through 12 at Cheshire Academy, plus its faculty, including librarian and technology coordinator George Aiello and French teacher Miriam Przybyla-Baum. More broadly, the piece speaks to teachers and administrators at any school still working out classroom AI policy, and to vendors like MagicSchool competing for that market alongside general-purpose chatbot makers such as Anthropic, Google and OpenAI, some of which are now building features aimed specifically at schools.

How to use it

MIT Technology Review's own recommendations: meet students where they are, since there is no way to fully police AI use on take-home work, so teach healthy habits rather than relying on bans; model best practices, because AI struggles with precision and voice and careless use by adults can read to students as permission to cut corners; and refine and replace, using AI to brainstorm lesson plans or problem sets, useful especially for early-career teachers without a large bank of material, while checking every citation, equation and statement before it reaches a final draft. Cheshire Academy's own mechanism is the traffic light label on each assignment: green for AI fully allowed, yellow for some tools such as spell-check but not others such as chatbots, red for none. Schools considering a dedicated platform can compare MagicSchool's free tier against its individual plan, priced just under $100 a year for unlimited access and full records.

How solid is it

This is a single case study built from what Cheshire Academy staff told MIT Technology Review for its Making AI Work newsletter, not a study with a comparison group or measured outcomes. The scale of AI adoption among teachers is described only in qualitative terms, a vast majority, many, some, with no breakdown by tool or by how much of the described program has been fully rolled out versus is still piloting, such as the Student AI Council. Whether the traffic light system or the student council improves learning, or whether either has been adopted anywhere beyond this one roughly 400-student private boarding school, is not addressed.

Risks and caveats

What works at a well-resourced private boarding school of about 400 students may not transfer to larger or less-resourced public schools facing the same policy questions. The school's own guidance flags that AI still hallucinates and struggles with precision and voice, so material generated for lesson plans or problem sets needs every citation, equation and statement checked before use. Enforcement of the traffic light system still depends on trust, since the article states plainly that there is no way to fully police AI use on take-home assignments. And MagicSchool's individual-plan price, just under $100 a year, is the only pricing figure given; the source names no institutional or school-wide license cost and no price for the platform's free tier.