Two hundred students from 39 schools across 19 states gathered in December for a deliberation on AI in education. Their consensus wasn’t what you might expect from a generation supposedly addicted to shortcuts: schools are giving them AI tools before teaching them to think.
“When a student turns to ChatGPT the moment a task feels difficult, that struggle gets bypassed,” the report notes. The students argued that foundational skills in reading, reasoning, and problem-solving should come first - potentially delaying general AI exposure until 9th grade.
This isn’t the narrative we usually hear about Gen Z and AI. But it aligns with what’s happening in classrooms around the world, where educators are wrestling with tools that outpace thoughtful policy.
The UAE Goes First
The UAE Ministry of Education issued the most comprehensive school AI regulations to date. Their Safe and Responsible Use of Artificial Intelligence in Classrooms 2026 guide introduces 25 prohibitions on generative AI use.
The headline rule: students under 13 and those below Grade 7 are barred from AI tools entirely. The Ministry cited “age-appropriateness standards” and concerns about “behavioral and educational impacts.”
Other prohibitions include:
- Submitting AI-generated work without disclosure
- Using AI during any exams or assessments
- Rephrasing AI-generated text without demonstrating genuine understanding
The rules acknowledge what many schools dance around: “Full reliance on AI to produce work undermines authentic academic effort.”
Whether these rules are enforceable is another question. But the UAE is at least drawing clear lines, which is more than most US school districts have managed.
The Analog Classroom
In Fort Worth, Texas, high school English teacher went almost entirely analog to keep generative AI out of her American literature and composition classes.
She teaches at Southwest High School, where most students come from low-income backgrounds. Her rationale is practical: these students need to develop the reading and writing skills that will serve them regardless of what AI can do. Handing them a tool that bypasses that development doesn’t help them.
About 60% of surveyed teachers report using AI in their classrooms at least occasionally. But the approaches are wildly inconsistent. Students in the December deliberation described policies that were “too strict, too permissive, inconsistent across teachers or disciplines, or unevenly enforced.”
One student’s perspective shifted after learning about a color-coded system clarifying appropriate AI use levels. The structure helped, not because it was permissive or restrictive, but because it was clear.
The Detection Problem
Schools that try to police AI use face a fundamental challenge: the detection tools don’t work as advertised.
GPTZero and Turnitin both claim around 99% accuracy, but those numbers come with asterisks. That accuracy only applies to unedited AI output. When students run text through humanizer tools or edit manually, accuracy drops below 20%.
The real-world numbers are worse:
- Independent studies find 3-4% false positive rates for native English speakers
- A Stanford study found detectors falsely flagged 61% of TOEFL essays written by non-native English speakers
- Turnitin’s Chief Product Officer acknowledged they catch “about 85%” of AI writing while trying to keep false positives under 1%
That 1% false positive rate means roughly one in every hundred human-written essays gets flagged. In a school with 2,000 students submitting multiple essays per term, that’s hundreds of false accusations per year.
For ESL students, the problem is worse. The tools were trained primarily on native English patterns, so non-native writing gets flagged at much higher rates. One 2026 assessment found Turnitin’s false positive rate spikes to 7% for ESL students.
When AI Tutoring Actually Works
The research on AI tutoring tells a different story. A randomized controlled trial published in Scientific Reports found students learn significantly more in less time when using an AI tutor compared with in-class active learning. They also reported feeling more engaged and motivated.
The key word is “tutor,” not “substitute teacher.” The successful implementations use AI for quick, focused practice and immediate feedback - not to replace human instruction entirely.
A systematic review of AI-driven intelligent tutoring systems in K-12 education found the most effective systems share common traits: they adapt in real time using interaction data, they adjust difficulty based on demonstrated mastery, and they complement rather than replace teacher-led learning.
The teacher’s role evolves from delivering standardized lessons to orchestrating learning experiences. AI dashboards show which students have mastered concepts and which need extra support. The teacher then focuses time where it matters most.
This is fundamentally different from handing students ChatGPT and hoping for the best.
The Skill Gap Problem
The students in the December deliberation identified the core issue: AI tools amplify existing skills but don’t create them.
A student who can write well can use AI to write faster. A student who can’t write doesn’t learn by watching AI do it. The cognitive struggle that builds understanding gets outsourced.
Their recommendations were specific:
- Delay general AI exposure, potentially until 9th grade
- Use education-specific AI tools with teacher-controlled parameters instead of general-purpose models
- Implement structured AI literacy education covering cognitive, ethical, and environmental implications
- Involve students in school AI policy decisions
The last point matters. These students want a voice in policies that affect their education. They’re not asking to be protected from AI - they’re asking for sequenced learning that prepares them to use it well.
What This Means
The AI-in-education debate has mostly been framed as permissive versus restrictive. But the students are pointing at something else: the order matters.
Foundational skills first, AI tools second. Clear policies over inconsistent ones. Human instruction supplemented by AI tutoring, not replaced by it.
The UAE’s blanket ban on AI for under-13s is blunt, but it acknowledges developmental readiness. The Fort Worth teacher’s analog classroom protects students who need more time to build skills. The successful AI tutoring programs work because they’re designed around learning, not around the AI.
Schools that hand students ChatGPT before teaching critical thinking skills are creating a dependency problem. The tools get better at writing while the students don’t.
What You Can Do
For parents: Ask what your school’s AI policy actually is. If teachers can’t explain it clearly, that’s a problem. Push for structured approaches rather than blanket bans or blanket permission.
For teachers: Consider sequencing. Students need to struggle with writing before AI helps them write. The cognitive load isn’t a bug - it’s the learning.
For students: The AI can write your essay, but it can’t make you smarter. The skills you build now will determine how effectively you can use these tools when they’re unavoidable.
For administrators: The students in the December deliberation asked to be included in policy discussions. They have perspectives worth hearing.