Last month, 98 student delegates from every U.S. state gathered at America’s Youth AI Festival on the UMass Boston and MIT campuses. They spent July 17-19 in workshops, discussions, and caucuses. They read policy proposals. They debated implementation. They drafted a national framework for AI in K-12 education. Then they voted on it. It passed 82 to 16.
The framework, the STUDENTS FIRST Act of 2026, is sharp and specific in ways that matter. AI should not generate artistic assignments. AI should not generate written work. Starting in ninth grade, students can use AI as a supplementary aid: for brainstorming when they’re stuck, for studying material they’ve already learned, for editing work they’ve already written. But only with teacher permission. And if a student gets accused of improper AI use, they get the right to appeal and demand meaningful human review. Schools cannot use AI to automatically profile students, assign grades, or make discipline decisions without human judgment in the loop.
Read those requirements together. The students were protecting three things: authentic thinking (you do the work), human judgment (someone makes the call), and relationships (the teacher knows the student).
It was a thoughtful framework.
One week later, OpenAI announced three new education plugins for ChatGPT Work and Codex. One was for K-12 teachers.
The plugin works like this: You select it. It presents you with a menu of teaching tasks. Create a lesson plan. Design an exit ticket. Write differentiated practice problems. Generate discussion questions. Build a formative assessment. The plugin doesn’t just suggest; it generates. It builds the lesson, structured and ready to use. It connects to Learning Commons to align everything with state standards and your district’s curriculum pathway. You customize the title and maybe swap one activity for another. Then you’ve got a resource packet, ready for tomorrow.
This is efficient design. Teachers have been asking for faster for years. We’ve been asking for tools that handle the heavy lifting, not tools that sit in the gap and make us do all the work. OpenAI solved for that. The press release said “Help teachers create engaging lessons faster.” The plugin does. A teacher could generate five different versions of a lesson in the time it used to take to make one.
But the collision is this: the student framework said no to AI-generated assignments. The vendor plugin is built to generate exactly those assignments. Students said AI in schools should be supplementary, supervised, limited. The plugin’s value proposition is that AI handles the generation workflow, making it the central tool, not the supplement. The plugin isn’t what students called for.
Nobody’s lying. Nobody violated the student framework. The framework isn’t law; it’s a policy proposal written by 98 high school students. But it reveals a gap: students have thought about what AI should do in schools. Teachers are getting tools built on a completely different assumption about what AI should do.
I’m not against the plugin or against fast lesson planning. Differentiation at scale is hard. If a tool can generate three versions of a lesson for different reading levels, that addresses a real need. But the job is this: notice the gap between what students are saying about this technology and what vendors are shipping.
The students said: protect authentic learning. That means students should do the writing, solve the problem, make the art. AI can help you prepare and gather resources, but it shouldn’t replace the actual work. In an English class, students write essays. In a history class, students construct arguments from evidence. In an art class, students make art. The AI can help them brainstorm what to write, but not write it.
The plugin says: here’s how to generate those assignments in two minutes.
The students said: require human judgment. That means the teacher decides when and how AI gets used in the classroom. The teacher is the final call. Not the curriculum designer, not the platform, not the algorithm, but the person who knows these specific students, in this specific class, on this specific day.
The plugin says: here’s how to make AI the default in your workflow.
One sentence from the framework stays with me: “Preserve the right of high school students and educators to decline AI use in appropriate circumstances.” Not opt-in, where you have to ask permission to skip the tool. Opt-out, where you have to ask permission to use it. Default to human decision-making, not to automation.
That’s the tension. Students drafted a framework that puts teachers back in control. Vendors are shipping tools that work best when teachers don’t have to decide; when the tool decides for them.
I notice that the students aren’t wrong, and the vendors aren’t wrong. They’re just optimizing for different things. The students are thinking about what school should be: a place where you do the thinking, where adults know you and make judgments about you, where relationships matter. The vendors are thinking about what tools can do: save time, reduce load, automate the routine.
For teachers, the job is to hold both in mind. You can use the OpenAI plugin. It can save you time on resource generation. That’s real. And you can use the STUDENTS FIRST framework to ask yourself: Am I using this because it serves my students’ learning, or because it’s fast? Does this assignment still require authentic thinking and original work, or have I automated away the actual task? Am I deciding when and how AI shows up, or is the tool’s design deciding for me?
The student framework gives you a checklist. “Does this protect authentic learning? Does this require my judgment? Does this serve the relationships that matter in my classroom?”
Those aren’t anti-AI questions. They’re pro-teaching questions.
Some of what the students wrote matters more than it first appears. “Require meaningful human review” of AI-related discipline. That sounds procedural. But it means: a student can’t get detention because an AI flagged their essay for plagiarism. A human has to read it. That’s teaching, not automation. It means consequences stay personal.
“Preserve the right to decline AI use.” That sounds like opting out. But it means: you can say no. You can run a classroom without the tool. You can be the teacher who decides that this assignment, this year, this group of students: this should be human-only.
“Prohibit schools from using AI to independently determine grades, discipline, or hiring.” That’s the core of it. The algorithm informs you; you decide. Your judgment doesn’t disappear into automation. Your judgment is the call. That’s what keeps school from becoming a system where nobody has to know anyone.
The students were speaking to us. Eighty-two of them voted yes on their framework. They said: AI can come to school. We get that. And we’re telling you how. It can supplement our work. It can’t replace it. We should decide how it shows up. You should decide how it shows up. Humans should stay in the loop.
That doesn’t mean you can’t use a tool that generates lesson plans fast. It means you can’t un-hear what the students said about why speed isn’t the same as good.
Takeaway for Teachers: Read the STUDENTS FIRST Act of 2026 framework (it’s publicly available on the AASA website). It’s not law, but it’s useful. Use it as a checklist before you adopt any new AI tool in your classroom. Does this assignment still require your students to do authentic thinking? Does your judgment stay in the decision-making? Can you decline to use it if it doesn’t serve your class? If the answer to any of those is no, that’s important data.
If you’re thinking through questions about AI’s role in teaching and learning, my book goes deeper. The AI Doesn’t Know Your Students is available on Amazon (https://www.amazon.com/dp/B0GXTNMBQV) and at shouldiuse.ai/book (https://shouldiuse.ai/book/).
David Jacobson is a high school history teacher. He writes about AI, education, and the messy intersection of the two at shouldiuse.ai.
