Oct 1, 2026
AI-Resistant Assignments for Middle and High School: Ideas by Subject
Nothing a student takes home is AI-proof. Here are AI-resistant assignment ideas for ELA, math, science, social studies, world languages, computer science and the arts, plus where in-class work still wins.
Jemel Fanfan | Founding Member at Flint
The question I hear most from middle and high school teachers about AI is what to assign now that a chatbot can do most homework in a few seconds.
They have reason to ask. In an NPR/Ipsos poll of 545 K-12 teachers (fielded April 27 to May 5, 2026), 57% said AI makes it harder to assess what students know. 39% said they're requiring more work to be done by hand, and 39% more work to be done in class. Students' use is growing at the same time. RAND found that the share of students using AI for homework rose from 48% to 62% between May and December 2025, driven mostly by middle and high schoolers.
In online teacher forums, the same two questions come up every few weeks: whether AI-proof homework exists, and whether it's time to switch to handwritten essays. The answers usually offered are put everything on paper, or give up. This post covers the space between those two: assignment designs for seven subjects, where handwritten work is still the best choice, and how to build AI into a task on purpose.
Can any assignment be AI-proof?
Not if students do it outside class. Any task a student completes alone at home can be handed to a chatbot, and some will. Only supervised work is close to AI-proof.
An AI-resistant assignment is one where the grade depends on thinking the student has to show. That might be data they collected, a decision they can explain out loud, a draft that changed in ways they can account for, or something said in your room that no chatbot heard. AI can still help with parts of the task, but it can't produce the evidence you're grading.
Most of the designs below use one of four moves:
- Tie the work to something only your class has: a discussion, a lab, a local place, a classmate's draft.
- Grade the process along with the product: drafts, checkpoints, revision notes.
- Add a short live check, like a two-minute conference, an exit ticket or a quick explanation at the board.
- Shift grading weight so that take-home work counts as practice and the in-class piece counts as the assessment.
AI-resistant assignment ideas by subject
These are starting points. Each one assumes students may use AI at home, and works anyway.
English language arts
| Assignment | Why it holds up |
|---|---|
| Respond to your class's own Socratic seminar: quote two classmates by name and push back on one of them | The chatbot wasn't in the room and doesn't know what was said |
| Draft in class, revise at home, then turn in a revision memo naming your three biggest changes and why you made them | You grade the memo and compare it to the in-class draft. Changes a student can't explain stand out |
| A three-minute writing conference: the student reads one paragraph aloud and explains one choice | A student who wrote it can talk about it. One who didn't usually can't |
Math
| Assignment | Why it holds up |
|---|---|
| Homework is practice, and a two-question in-class check the next day carries the grade | Copying answers at home stops paying off when the grade comes from what they can do on their own |
| Collect your own data (commute times, sleep, steps) and fit a model to it, with a photo or log of the raw numbers | The data belongs to the student, and the reasoning has to fit it |
| Write a problem this method solves, then one it can't solve, and explain the difference | Building a problem shows understanding in a way solving one doesn't |
Science
| Assignment | Why it holds up |
|---|---|
| Lab conclusions must cite your group's measured values and explain one result that didn't match the prediction | A chatbot can write a generic conclusion. It can't explain your group's outlier |
| Predict, observe, explain: students write a prediction in class before a demo, then explain the gap afterward | The prediction is on paper, timestamped, before anyone sees the answer |
| Peer-review another group's lab report in class, using a short rubric | Critiquing a specific report means reading it closely |
Social studies
| Assignment | Why it holds up |
|---|---|
| Interview a family member or neighbor about a historical event they lived through, and connect it to a primary source from class | The interview is original material, and the connection has to be argued |
| Prep at home, debate in class: students research both sides, then draw their position on the day | Students can use AI to prepare, but the grade is for the live argument |
| A local history project using your town's archive, historical society or a building students can visit | Local, specific sources are thin online, so students have to go and look |
World languages
| Assignment | Why it holds up |
|---|---|
| Unrehearsed conversation in pairs or with you, on a prompt students see on the day | Translation tools are no help in real time, face to face |
| Voice memo answers to a question revealed only when recording starts | There's no time to write a script and read it |
| Homework is listening and reading, and writing happens in class | AI is strongest at producing text, so the text gets produced where you can see it |
Computer science
| Assignment | Why it holds up |
|---|---|
| A two-minute code walkthrough: the student explains their program line by line, then makes a small change you ask for on the spot | Explaining and changing code you didn't write is hard to fake |
| Grade the commit history or saved checkpoints along with the final program | Working code that appears all at once in a single commit is a signal worth asking about |
| Debug a classmate's program and write up what was wrong and how you found it | The bug is specific to that program, and the write-up shows the search |
The arts
| Assignment | Why it holds up |
|---|---|
| A process portfolio or sketchbook with dated stages, plus an artist statement tied to specific decisions | Grading the stages makes the making visible |
| In-class critique, where students describe a peer's piece using the vocabulary you've taught | It happens live, about work in the room |
| Music: a recorded performance, or a composition submitted with two earlier drafts | Performance is live by nature, and drafts show how a composition grew |
When is handwritten or in-class work the right call?
When the grade has to mean the student did it alone. That covers first drafts you need to see unassisted, short checks of what a student knows, and anything you'd treat as a test.
That's level 1 on the AI Assessment Scale, a free framework from researchers Mike Perkins, Jasper Roe, Leon Furze and Jason MacVaugh. Level 1 is the only one of its five levels that calls for supervised conditions. The others describe tasks where some AI use is expected and planned for.
In-class work has costs, too. It uses class time you'd otherwise spend teaching, and handwriting can penalize students who write slowly or whose accommodations call for typing. Keep IEP and 504 accommodations in place for paper-based work, and save supervised work for the assessments where it matters.
Can AI be part of the assignment?
Yes, and for some skills it should be. Students will use these tools after they leave your class, and judging AI output is a skill you can teach and grade.
A few designs that build AI into the task:
- Fact-check the chatbot. Give students an AI-written summary of a topic you've taught, with errors in it. They find the errors and cite the evidence that proves each one wrong.
- Argue with it. Students ask a chatbot for an argument, then write a rebuttal using sources from class. You grade the rebuttal.
- Show the conversation. Students turn in their chat log alongside the work, with a short note on which suggestions they used, which they rejected, and why.
The last one changes what you're grading: you read how the student used AI and grade their judgment about it. This matches level 3 on the AI Assessment Scale, where the task is designed so that "AI alone will not reach the required standard."
Whichever level you choose, write it on the assignment. RAND found that only 45% of students consider getting direct answers to homework from AI to be cheating, and only about a third said their school had a schoolwide AI policy. Students can't follow a rule nobody stated. A traffic light poster works well for younger students.
What about AI detectors?
I wouldn't build a plan around them. Turnitin, whose detector many schools use, says in its own guidance that its AI writing model "may not always be accurate" and "should not be used as the sole basis for adverse actions against a student." Our CEO wrote in 2023 about why we think AI writing detectors don't work.
The designs above give you better evidence than a percentage: drafts, checkpoints, and a student who can explain their own work.
Where Flint fits
I work at Flint, so read this section knowing that. Flint is an AI learning platform where teachers create activities and students work through them with Sparky, Flint's AI tutor.
If you try the "show the conversation" approach, Flint does that part for you. Teachers can read every conversation in the activities they create, and they can see any text a student pasted in, so the process is there to read without collecting chat logs. For how to set up an activity like this, see 4 tips for creating a successful AI assignment, or read about six classroom-tested activities from McDonogh School.
FAQ
What is an AI-resistant assignment?
An AI-resistant assignment is one where the grade depends on thinking the student has to show: their own data, a choice they can explain, a draft that changed, or a conversation in class. A chatbot can still help, but it can't produce the evidence the teacher is grading.
Can an assignment be AI-proof?
Not if students do it outside class. Any task a student completes alone at home can be handed to a chatbot. Only supervised work is close to AI-proof, so the realistic goal for homework is AI-resistant.
Should I make every assignment handwritten?
No. Handwritten, in-class work is the right call when the grade has to mean the student did it alone, like a first draft or a short check of what they know. Use it for those moments, keep accommodations in place for students who type, and design the rest so the thinking is visible.
Do AI detectors work?
Not well enough to rely on. Turnitin's own guidance says its AI writing model may not always be accurate and should not be the sole basis for adverse actions against a student. Drafts, process and a conversation with the student are better evidence.
How do I tell students what AI use is allowed on an assignment?
Name a level on every assignment, from no AI to AI as part of the task. The AI Assessment Scale gives five levels you can borrow, and a red, yellow and green poster works for younger students. For more on teaching with AI, see our free AI literacy course for teachers.