Somewhere between the enterprise pilots and the funding rounds, AI quietly became a school subject. Not an elective, not a club — a required one, with a minimum hour count written into policy. Beijing got there first at scale, and this year the requirement stops being a big-city experiment and becomes a national floor across China. The United Arab Emirates already teaches it from kindergarten. India starts next academic year at class 3.
This matters beyond education policy, because it is the clearest signal yet of where AI is heading as a category. Governments do not mandate a subject because a technology is exciting. They mandate it when they have decided the technology is infrastructure — something every citizen will have to operate, like reading a timetable or using a bank account. Here is what has actually been mandated, what is being taught in those hours, the guardrails that got much less coverage than the headlines, and what it means for the tools people actually use.
The short version: who requires what
| Where | What is required | From when |
|---|---|---|
| Beijing 🇨🇳 | Compulsory AI education across every grade, no fewer than 8 class hours, in 1,400+ primary and secondary schools | September 2025 semester |
| Hangzhou 🇨🇳 | Compulsory AI course, at least 10 class hours per year | 2025-26 |
| China (national) 🇨🇳 | Every elementary and middle school must offer an introductory AI course; coverage "basically universal" at K-12 by 2030 | 2026-27 academic year |
| UAE 🇦🇪 | AI as a compulsory subject from kindergarten to grade 12, seven subject areas, no exams | 2025-26 academic year |
| India 🇮🇳 | AI and computational thinking compulsory from class 3 (CBSE framework built with IIT Madras, under NEP 2020) | 2026-27 academic year |
| United States 🇺🇸 | No mandate. Executive Order 14277 creates a task force, teacher training and a Presidential AI Challenge | Signed 23 April 2025 |
| Estonia 🇪🇪 | No standalone AI subject; AI integrated into existing curriculum from grade 7 | Ongoing |
Two things jump out of that table. The first is how small the hour counts are: eight to ten class hours a year is roughly one lesson a month. The second is that the countries moving fastest are not the ones with the biggest AI labs. They are the ones that treat curriculum as an instrument of industrial policy.
What Beijing actually did
Beijing became the first provincial-level region in China to launch compulsory, comprehensive AI education for primary and secondary schools. According to China Daily, over 1,400 primary and secondary schools in the city provide no fewer than eight class hours of AI education covering all grade levels from primary through senior high, starting with the semester that began in September 2025. In practice that makes AI literacy mandatory from age six.
The interesting part is what fills those hours, because it is not a lecture about neural networks. China Daily lists the classroom stack as block-based programming platforms, iFlytek robots with image recognition, and large language models that help students generate executable code from instructions written in Chinese. That last one is worth pausing on: an eleven-year-old writing a plain-language instruction and watching a model turn it into code is doing, at primary school, exactly what the rest of us discovered in 2023.
At Haidian District Experimental Primary School, grades one to six get project-based lessons that escalate year by year. Shi Yuan, who heads the school's information centre, described the design goal in a way that would not be out of place in a product brief: "We turn abstract concepts into tangible projects. It's not about making every child an expert, but about helping each one understand and interact confidently with AI."
Guangqumen Middle School runs a continuous curriculum spanning primary, middle and high school. Seventh graders take a semester-long information technology course; eighth and ninth graders design "intelligent study companions" using generative AI; the high school AI class, built with partners including the Chinese Academy of Sciences and Huawei, reaches into embodied intelligence, brain-computer interfaces and autonomous driving. Students there have operated a robotic arm using a smart headband in class. Pei Yi, the school's director of talent cultivation, framed the point of all of it: "In the future, human-machine collaboration will be inevitable. Students may not all become AI specialists, but they must learn how to use AI effectively."
Hangzhou — home to DeepSeek and to Unitree Robotics — went further than the capital, making AI classes compulsory with a floor of at least ten class hours a year, taught either as an intensive week, folded into information technology and science lessons, or run through after-school programmes.
The guardrails nobody reported
The coverage of China's push was almost entirely about the mandate. The rules that came with it are arguably more interesting, because they are stricter than what most Western schools have written down.
China's Ministry of Education issued guidance on strengthening AI education in primary and secondary schools in December 2024, followed by further guidelines in May 2025. Under them, primary school students are prohibited from independently using open-ended content generation tools. Teachers are barred from using generative AI as a substitute for their primary teaching role, from relying on AI to answer students' questions directly, and from using AI-generated content to evaluate students. Feeding sensitive data — personal information, exam questions — into AI tools is strictly forbidden. And there is no plan for standardised national AI testing before 2030.
China mandated the subject and then banned the most obvious way of teaching it. The policy is "learn how this works", not "let the chatbot do your homework" — and the distinction is the entire ballgame.
That combination — compulsory exposure, restricted autonomous use, no exams — is a defensible answer to a question every education system is currently fumbling. It also means the mandate is not, in itself, a growth market for consumer chatbots in classrooms. It is a growth market for structured, supervised, curriculum-shaped tools.
The American version: incentives instead of hours
The United States is chasing the same outcome with a completely different instrument, because federal government cannot set curriculum — that is a district-by-district decision. Executive Order 14277, "Advancing Artificial Intelligence Education for American Youth," was signed on 23 April 2025. It created a White House Task Force on AI Education, directed the Department of Education, the National Science Foundation and the Department of Labor to build K-12 resources and teacher training, and launched a nationwide Presidential AI Challenge, with agencies working to a July 2026 deadline for that launch.
No hour count, no required subject, no age floor. Whether that is nimbler or slower than a mandate is exactly the experiment now running, and we will not know the answer for years. What we do know is that the American classroom is already full of AI — it just arrived through teachers rather than through policy.
The real adoption story is the teachers, not the students
This is the number that reframes the whole topic. In the Gallup and Walton Family Foundation study "Teaching for Tomorrow: Unlocking Six Weeks a Year With AI" — a survey of 2,232 US public K-12 teachers conducted between 18 March and 11 April 2025 — teachers who used AI tools at least weekly reported saving an average of 5.9 hours a week. Across a school year, that is roughly six weeks of time returned.
The adoption split in the same survey is just as telling: about 32% of teachers used AI at least weekly, 28% used it infrequently, and 40% were not using it at all. So the picture in mid-2025 America was a third of teachers quietly reclaiming a month and a half of their year, and two-fifths not touching the technology.
That gap is the real story of AI in education right now. Students are getting eight mandated hours a year in Beijing. Teachers are getting six weeks back — voluntarily, unevenly, and with no policy telling them to.
What tools this actually runs on
Strip away the policy language and the classroom stack falls into four jobs, none of which needs a frontier model:
- Lesson and material generation. The single biggest teacher use case: turning a curriculum standard into a lesson plan, a worksheet, a differentiated reading at three levels. This is where the time savings in the Gallup data come from, and it is dominated by education-specific tools rather than raw chatbots, because the value is in the templates and the guardrails.
- Tutoring and practice. Guided, subject-bounded tutors that walk a student toward an answer instead of handing one over. Khanmigo is the best-known example of the pattern; the design constraint is Socratic behaviour, not raw capability.
- Comprehension and research support. Explaining a dense passage, summarising a source, checking a paraphrase. This is the category students adopt on their own, with or without permission — the tools in our best AI tools for students roundup are mostly this.
- The physical layer. The part of the Chinese curriculum that has no Western equivalent yet: block programming driving actual hardware, image-recognition robots, robotic arms. It is the same shift we wrote about in how AI is moving into physical form, arriving in classrooms before it arrives in homes.
If you are a teacher or a parent choosing between them, the honest advice is to pick by job rather than by brand. A tutor that refuses to give the answer is a different product from an assistant that writes your lesson plan, and both are different from a general-purpose chatbot with a school-shaped prompt in front of it.
What "AI literacy" should actually mean
The risk in every one of these programmes is that "AI education" collapses into prompt tricks — a skill with roughly the shelf life of knowing the right search operators in 2009. The curricula that look most durable teach three things that do not expire:
- What the machine is doing. That a model predicts rather than knows, that it will be confidently wrong, and that a plausible sentence is not evidence. China's tiered approach puts model training fundamentals and AI ethics at secondary level for exactly this reason.
- Where it fits in your own work. Not "can AI do this" but "should it, here, given what I am trying to learn". The eighth graders at Guangqumen designing their own study companions are doing a better version of this than most corporate training.
- What it costs and who sees your data. The most practical adult AI skill, and the one no school curriculum currently covers well. It is also why China's rules explicitly forbid putting personal information or exam questions into AI tools.
The honest caveat
None of this is proven yet. There is no evidence, because there cannot be evidence, that a mandated eight hours a year produces more capable adults — the first cohort taught under Beijing's rules is currently in primary school. China has explicitly declined to test the subject before 2030, which is either admirable restraint or a way of avoiding an inconvenient measurement, depending on how cynical you are feeling.
What is certain is the direction. Four of the world's largest school systems have decided that not teaching AI is riskier than teaching it badly, and they are moving on that basis. Within a decade, the default assumption behind every AI product will shift: from users who need to be persuaded that a chatbot is useful, to users who were taught what one is before they were taught long division.
Frequently asked questions
Is AI education mandatory in schools?
In some places, yes. Beijing made comprehensive AI education compulsory across all grades in more than 1,400 primary and secondary schools starting with the September 2025 semester, with no fewer than eight class hours. Hangzhou requires at least ten hours a year. The UAE made AI a compulsory subject from kindergarten to grade 12 in the 2025-26 academic year, and India makes AI and computational thinking compulsory from class 3 in 2026-27. China's Ministry of Education requires every elementary and middle school nationwide to offer an introductory AI course from the 2026-27 academic year.
How many hours of AI do students actually get?
Far fewer than the headlines suggest. Beijing's floor is eight class hours and Hangzhou's is ten per year — roughly one lesson a month. The mandates are about exposure and literacy, not about producing engineers.
Are students allowed to use ChatGPT-style tools in class?
Under China's guidelines, primary school students are prohibited from independently using open-ended content generation tools, and teachers may not use generative AI as a substitute for their core teaching role or to evaluate students. Elsewhere the rules are set school by school. The consistent pattern in policy is supervised, bounded use rather than open access.
Which AI tools do teachers use most?
The heaviest use is not tutoring — it is preparation: lesson plans, worksheets and differentiated materials. In the Gallup and Walton Family Foundation survey of 2,232 US public K-12 teachers, those using AI at least weekly saved an average of 5.9 hours a week, about six weeks over a school year, while 40% were not using AI at all.
Does the US require AI education?
No. Executive Order 14277, signed on 23 April 2025, creates a White House Task Force on AI Education, teacher training programmes and a nationwide Presidential AI Challenge, but sets no required hours or subjects — US curriculum is decided at district level.
What should a student learn about AI that will still matter in ten years?
How the model actually works and why it can be confidently wrong; when using it helps your learning versus when it replaces it; and what it costs and who sees your data. Prompt technique changes with every model release. Those three do not.