Eighty percent of teachers across Europe worry that artificial intelligence has entered classrooms faster than anyone can manage. Students already turn to chatbots for homework. Many do so weekly. The gap between what educators see and what they can control grows wider by the month.
A fresh survey from Epson lays out the mismatch in stark numbers. Sixty-eight percent of teachers say AI-assisted homework harms actual learning. Yet 76 percent of students expect to use the technology as part of their studies. Almost 90 percent already reach for it once a week. And 81 percent of teachers believe their pupils think AI can handle spelling and basic math without effort. (TechRadar, Sept. 12, 2026)
These figures arrive at a moment when schools on both sides of the Atlantic scramble for answers. In the United States, an NPR/Ipsos poll of 545 K-12 teachers found 54 percent believe AI makes it harder for students to develop critical thinking. Fifty-five percent call it mainly a shortcut to avoid real work. Only 33 percent said their school has a formal policy on student use. (WINSSolutions, reporting June 2026 NPR/Ipsos data)
But the story runs deeper than numbers. Teachers don’t just fear student overreliance. They also lack training to guide the tools responsibly. The Epson research shows 82 percent want more preparation so they can supervise AI use effectively. Without it, many feel powerless.
Policy lags behind adoption
Some districts have moved faster than others. New York City recently placed a one-year moratorium on student-facing generative AI tools through eighth grade and limited screens in early grades. More than 1,100 parents in Bend, Oregon, signed a petition earlier this year to remove such tools from devices. Fairplay, a children’s advocacy group, called for a five-year pause on generative AI products aimed at students from preschool through high school. (The Guardian, June 23, 2026)
Parents and experts voice a common complaint. Ed tech companies appear to set the direction while teachers become technical support. Kelly Clancy, a parent in Brooklyn with three children in public schools, described an assignment that asked her sixth-grader to build a science experiment and then consult Google Gemini for feedback. She pushed back. The chatbot, she argued, teaches children that machines can do the thinking. Better to discuss improvements with classmates.
Similar unease shows up in Canada. A recent study found 96 percent of teachers believe at least some students use AI for homework. More than 42 percent think over half do. Nearly 56 percent of those users reportedly let the technology complete the work entirely. Eighty-four percent of Canadian educators say students’ critical thinking skills have declined compared with five years ago. (Toronto Sun, Aug. 27, 2026)
Yet not every educator sees only danger. Some experiment with new approaches. At the University of Baltimore, Rachael Zeleny, director of the writing program, urged instructors to adopt a free tool called Brisk. It helps spot AI-generated work quickly so teachers avoid wasting time on machine-produced essays. The real shift, however, goes beyond detection. Many now redesign assignments to demand evidence of genuine learning. In-class writing. Oral defenses. Projects built step by step under supervision. (The Washington Post, Sept. 6, 2026)
These changes reflect a broader realization. Traditional take-home essays no longer suffice as proof of mastery. AI can produce fluent prose on demand. The question becomes how to measure understanding when machines generate text that sounds better than many student efforts.
Even teachers who use AI themselves draw firm lines. A study published in PNAS Nexus revealed a surprising pattern. Educators claim they distrust AI grading systems. They rate them less fair and less competent than human colleagues. Yet when presented with overly harsh AI grades, they correct them less often than identical errors from humans. The gap between stated skepticism and actual behavior stood out to researchers. Many still rely on the systems for lesson planning. Nearly half use generative tools weekly for that purpose. (PsyPost, June 28, 2026)
The pattern repeats in other countries. In India, over 70 percent of teachers now use AI tools, according to a CENTA survey released this week. Lesson planning tops the list of applications. Yet 84 percent express worries. Accuracy problems and fears that AI could replace human roles rank high. Only a minority have received structured training. Many enter student data into public AI systems, raising privacy risks. (This Week India, Sept. 12, 2026)
And the technology keeps advancing. Early experiments with agentic AI systems that can act autonomously already worry researchers. These go beyond generating answers. They complete multi-step tasks, potentially handling entire assignments with little oversight. Schools that barely adapted to chatbots now face more sophisticated tools. A Brookings Institution study across 50 countries warned that unguided AI use can undermine cognitive, social and emotional growth. (Brooklyn Eagle, Sept. 8, 2026)
Some teachers respond by returning to analog methods. In New York, certain educators have ditched screens almost entirely. One Brooklyn veteran with 25 years of experience said he went back to his earliest teaching style after seeing students retain nothing from AI-generated answers. ChatGPT gave perfect homework results but produced failure on follow-up tests. The realization hit hard. The kids didn’t learn.
Others integrate the tools deliberately but with guardrails. They use AI to personalize explanations or generate practice questions while reserving human interaction for discussion and feedback. The goal stays clear. Keep the teacher at the center. Use technology to free time for relationships rather than replace them.
Research supports that instinct. A randomized trial in Turkey involving 193 teachers and over 2,800 students found that when instructors relied on an AI assistant for materials, their students reported lower motivation. The effect was strongest among weaker teachers to begin with. Final exam scores dropped. Lead researcher Alp Sungu of the Wharton School suggested many treat AI as a crutch instead of a partner. They generate lesson plans with minimal editing. Quality suffers. Students notice. (The Hechinger Report, July 13, 2026)
Universities and governments begin to answer the training gap. Neumann University in Pennsylvania will offer a state-approved AI endorsement for K-12 teachers starting next summer. The 12-credit online program covers leadership, ethical use, and designing learning experiences. Similar efforts appear elsewhere as policy catches up.
Yet the pace feels uneven. In the UK, three-quarters of secondary students say they want teachers to use AI to make hard topics clearer. At the same time, only 15 percent report receiving enough guidance on responsible use. Students crave the tools. Schools provide few rules. Teachers stand in the middle. (Deep Education Network, June 12, 2026)
The tension shows no sign of easing. New York lawmakers introduced the FOCUS Act to create a right to analog learning and sharply limit screen reliance. The American Federation of Teachers partnered with major AI companies on training programs even as it warns against early exposure for young children. No screens for second grade or younger. No chatbots in elementary school.
One thing seems certain. AI will not replace teachers. The human elements matter too much. Empathy. Context. The ability to notice when a student struggles for reasons beyond the lesson. Seventy-three percent of UK secondary students in one survey agreed AI cannot replicate those qualities. The mechanical tasks may shift. The relational core remains.
Still, the pressure builds. Schools must decide how to prepare students for a world where AI exists everywhere while protecting the habits of mind that no machine can instill. Teachers need support, clear policies and time to adapt. Without those, the race continues. And many feel they are losing ground.
Recent coverage shows the debate intensifying. A Korea Times column from this week described how AI could transform language learning through chatbots and virtual practice yet clashes with rigid exam systems that reward memorization over fluency. (The Korea Times, Sept. 10, 2026) Instructure’s fresh research, also published today, finds parents, students and educators share top worries: overreliance, loss of critical thinking and academic integrity. Confidence in AI remains low across groups. Training lags further behind. (Instructure, Sept. 12, 2026)
The data paints a consistent picture. Adoption races ahead. Understanding and oversight trail. Teachers see the risks clearly. Many already live with the consequences in their classrooms every day. The question now is whether institutions can catch up before another generation learns to outsource its thinking.