AI in the Classroom: How One School Stopped Panicking and Started Teaching With It
Cheshire Academy in Connecticut tried banning AI, then tried ignoring it. Neither worked. Here is what they are doing instead.
Key points
- Cheshire Academy, a private school in Connecticut with around 400 students, has introduced a traffic-light labelling system to tell students when AI is and is not allowed on assignments.
- The school's "Student AI Council" puts students in charge of leading discussions about healthy AI use.
- AI educator platform MagicSchool costs just under $100 per year for an individual teacher's unlimited plan.
- French teacher Miriam Przybyla-Baum has taught at Cheshire for nearly 30 years without using AI herself, yet runs some of the school's most creative AI-awareness exercises.
- AI tools can generate plausible but incorrect content, a failure mode researchers call hallucination, making human review of every output essential.
Chatbots hit schools like a surprise fire drill. One day students were writing their own essays; the next, they had an app on their phones that could produce a passable five-paragraph piece in about ten seconds. Teachers noticed. But knowing something is happening and knowing what to do about it are very different problems.
Cheshire Academy, a boarding and day school in Connecticut, decided the answer was not a blanket ban.
What did the school actually do?
Instead of picking one AI tool and mandating it, Cheshire trained staff in general techniques: how to write a useful prompt (a typed instruction that tells an AI what to do), and crucially, where AI goes wrong. Teachers now use a mix of general chatbots like ChatGPT and Perplexity alongside MagicSchool, a platform built specifically for educators that can generate quizzes, worksheets, grading rubrics and lesson plans from a single dashboard.
The traffic-light system is the most visible result. Every assignment now carries a colour code. Green means AI is fully allowed. Red bans it entirely. Yellow sits in the middle: a teacher might permit spell-check but not a full chatbot conversation. Students know the rules before they open their laptops.
For students, this is genuinely useful information. No more guessing whether their teacher minds a grammar suggestion or would see it as cheating.
What about the teachers who still do not want to use AI?
They are welcome to opt out, and some do. Miriam Przybyla-Baum teaches French. She has been building classroom materials for nearly 30 years and does not feel she needs a chatbot's help. But she spotted students using AI-assisted tools like Google Translate to dodge language exercises long before ChatGPT launched. So she built exercises around that reality.
In one assignment, students let a large language model (an AI system trained on vast amounts of text, the same technology behind ChatGPT) edit their homework. Then they go back through every suggested change and decide which ones improved their writing and which ones flattened their voice. In another exercise, students anonymously grade each other's AI-assisted work, flagging which sentences they suspect a machine wrote.
The point is not to catch cheaters. It is to make students think.
Should teachers trust AI to write for students?
Not without checking it carefully. AI systems frequently produce what experts call hallucinations: confident-sounding statements that are simply wrong. A generated quiz might contain a flawed equation or a mangled historical date. Any teacher using AI for student-facing materials needs to read every line before it goes out.
MIT Technology Review's reporting on the Cheshire case highlights a practical rule worth keeping: use AI to brainstorm and draft, then treat that draft the way you would treat a student's first attempt. It needs a human eye.
Common questions
Can a school just ban AI and move on?
Probably not for long. Students have access to these tools on personal devices at home, and enforcing a ban on take-home work is close to impossible. Teaching students when and how to use AI responsibly tends to produce better outcomes than prohibition.
Is MagicSchool safe to use for student data?
MagicSchool markets itself as an educator-first platform with privacy controls, but teachers at Cheshire have so far held back from using it to generate personalised student feedback, citing concerns about data privacy and whether the output would be good enough. If your school is considering any AI tool, ask the vendor directly for its data policy before uploading anything about students.
Does a teacher need to spend money to use AI in the classroom?
Not necessarily. Free tiers of ChatGPT and similar tools handle most planning and brainstorming tasks. MagicSchool also has a free version, though the paid plan, at just under $100 a year for one teacher, unlocks unlimited use and full record-keeping.
