Two Roads: What China and the U.S. Are Actually Doing With AI in Schools

A strange thing happens when you spend some time reading primary-source AI-in-education policy from both countries side by side. The question most American educators are still arguing about should AI be in schools is not the question Chinese policymakers are asking. They answered that one in 2017. The question they are now operationalizing is harder, and it is the one we should be asking too: how, at what scale, with what guardrails, by what date, and how do we fund the rural gap?

I want to be careful here, because comparisons across systems can flatten what is actually happening on the ground in both countries. Neither system is monolithic, and as I’ll get to below, the Chinese data in particular deserves more scrutiny than a first read suggests. But after pulling the primary documents the Chinese Ministry of Education’s twin guidelines from May 2025, the 2023 Generative AI Services Measures, the Trump administration’s April 2025 executive order on AI education, and the patchwork of state-level guidance now covering 34 U.S. states plus Puerto Rico a clear structural difference emerges that is worth naming plainly.

The Chinese Approach: Curriculum as the Core

China’s national AI-in-education push originates in a 2017 State Council document, the New Generation AI Development Plan (国务院, 2017), which explicitly mandated “setting artificial intelligence-related courses in primary and secondary schools” and “gradually promoting programming education.” That single sentence is the seed. Everything since has been layered on top of it.

The 2017 mandate was followed by curriculum revisions that embedded AI content into compulsory K-9 information technology standards. This is not aspirational. It is a national curriculum standard, and textbooks in every province are aligned to it.

By May 2025, the Ministry’s Basic Education Teaching Guidance Committee released two companion guidelines that operate at the classroom level: the AI Literacy Education Guide and the Generative AI Use Guide (教育部基础教育教学指导委员会, 2025). The Literacy Guide builds a four-stage spiral curriculum, with elementary school focused on interest and basic cognition, junior high on technical principles and basic application, and high school on systems thinking and innovative practice. The Use Guide is more prescriptive than anything I have seen from any U.S. state.

It forbids elementary students from using open-ended generative AI alone. It forbids students from submitting AI-generated content as homework or exam answers. It forbids teachers from using AI as a substitute teaching subject. It forbids inputting personal information or exam questions into AI tools. And it explicitly names the risk of cognitive offloading, the fear that students will lose independent thinking if AI does the thinking for them.

The hard-law instrument behind all of this is the 2023 Interim Measures for the Management of Generative AI Services (国家网信办 et al., 2023), a binding departmental regulation from seven ministries including the Cyberspace Administration of China and the Ministry of Education. Article 10 requires providers to take effective measures preventing minor users from over-relying on or becoming addicted to generative AI services. That is the only hard-law minor-addiction-prevention clause in any major jurisdiction worldwide, and it applies to every commercial edtech product serving Chinese schools.

China is also naming the rural gap explicitly. National programs are pushing resources toward less-developed regions, though the scale and effectiveness of these efforts deserve independent scrutiny.

A caveat worth sitting with: nearly every one of those figures comes from a government ministry or state media outlet. That matters. Independent, on-the-ground reporting tells a messier story. Journalist Lily Ottinger’s 2026 field reporting on China’s AI-education rollout (ChinaTalk, 2026) documents commercial vendors like iFlytek charging individual schools upward of $250,000 for full classroom AI suites, huge infrastructure variance between wealthy coastal counties and poor inland ones, and a documented case of a prominent Chinese high school misrepresenting its gaokao admissions data by bundling numbers with private satellite campuses to look better in provincial reporting. The article also warns that the same incentive structure local administrators evaluated on metrics they can manipulate will likely produce cherry-picked AI-education evaluation data as rollout proceeds. A national curriculum standard existing on paper and being delivered with fidelity in a Gansu township school are two different claims. Self-reported metrics inside a top-down accountability system carry a known incentive to look better than the reality, and I want to flag that rather than let the numbers above stand unquestioned.

The American Approach: Guidance, Patchwork, Aspiration

The U.S. picture is genuinely different in structure, and I want to describe it fairly.

At the federal level, the most significant recent move is the April 2025 executive order, “Advancing Artificial Intelligence Education for American Youth” (White House, 2025). It establishes a White House Task Force on AI Education, directs the Department of Education, NSF, and Department of Labor to collaborate on AI literacy and educator professional development, and launches a Presidential AI Challenge for student and educator innovation. In 2025, the Department of Education under Secretary Linda McMahon issued a Dear Colleague Letter allowing federal grant funds to be used for AI in education, with emphasis on parent and teacher engagement in ethical use (U.S. Department of Education, 2025).

But here is the structural difference: there is no federal K-12 AI curriculum standard. There is no binding national framework that says what a third grader or an eighth grader should know about AI. The 2022 National Educational Technology Plan from the prior administration offered a vision. It did not mandate content.

What exists instead is a state patchwork. As of mid-2025, 34 states and Puerto Rico have some form of official K-12 AI guidance or policy (AI for Education, 2025). California’s Department of Education issued comprehensive guidance. Alabama released an AI policy template for local education agencies in June 2024. Alaska, Arkansas, Georgia, Hawaiʻi, Maryland, Mississippi, and others have introduced AI-related education bills or formed task forces in 2025 (Education Commission of the States, 2025). The Southern Regional Education Board’s Commission on AI in Education and state task force reports from Arkansas and Georgia call for comprehensive risk-management policies, cross-sector collaboration, and phased policy development.

This patchwork has real strengths. It allows California to lead with a comprehensive approach while Mississippi experiments with a task force model. It respects local control, which is a genuine American value in education governance. It lets districts pilot and iterate.

It also has real costs. A student in a district with strong AI guidance gets systematic exposure to AI literacy, ethics, and tool use. A student in a district without it gets whatever their individual teacher decides to do, which, in many cases, is shallow. And here I need to be honest about what that same data actually shows.

Where the Two Systems Actually Diverge, and Where My Own Evidence Complicates the Story

Three differences matter most for practitioners, and I want to state the third one more carefully than I did the first time I wrote this.

First, curriculum. China has a binding K-9 national curriculum standard that embeds AI as mandatory content. The U.S. has no equivalent. State guidance is real and growing, but it is guidance, not curriculum. This is the single largest structural difference, and it cascades into everything else.

Second, guardrails. The Chinese 2025 双指南 (twin guidelines) prohibitions, no elementary solo generative AI use, no AI content submitted as homework, no AI as substitute teacher, no sensitive data input, are more prescriptive than any U.S. state policy I have reviewed. The U.S. conversation has been heavier on opportunity and lighter on prohibition. That is not necessarily wrong; American classrooms operate in a different rights framework. But practitioners should know the Chinese prohibitions exist because researchers and policymakers there identified specific cognitive and ethical risks worth naming directly (罗生全 et al., 2023; 王继新 & 黄柳苍, 2025).

Third, depth versus structure, and this is where I have to push back on my own thesis. Even inside a system with binding curriculum and mandated training, there is evidence that adoption depth remains uneven. That is not a footnote. It is a reminder that structure alone does not produce depth. If a national curriculum mandate and hard-law guardrails do not automatically produce deep classroom integration, then the honest conclusion is not simply “China is further along.” It is that curriculum and prohibition are necessary but not sufficient, and that the harder, slower work, building teacher capacity and trust at scale, is unsolved in both countries. I would rather say that plainly than let a clean structural story override the reality that complicates it.

What Both Systems Share

Both systems are wrestling with the same teacher-capacity bottleneck, and it is not a fair fight. Chinese teachers are working inside mandated curriculum and qualification-exam requirements, with national-level investment in education-domain AI infrastructure. American teachers are largely being asked to close the same depth gap with none of that scaffolding: no national curriculum to anchor to, no protected professional development time in most districts, and no education-specific tool handed to them by their state. When a U.S. teacher’s adoption looks shallow next to a Chinese teacher’s, the fair comparison is not “one is trying harder.” It is that one has been given a system to work inside and the other has been given a mandate with no system underneath it. If depth is the real bottleneck in both countries, as I believe it is, then the fix in the U.S. starts with funding and building that scaffolding, not with asking individual teachers to close the gap through effort alone.

Both systems are also investing in education-domain AI tools rather than relying solely on consumer chatbots. China has been building domestic-base education AI models through its universities and national programs. In the U.S., organizations like TeachingLab and the AI for Education consortium are working on AI integration in education and curating guidance for responsible classroom use, though neither is building a custom education-domain LLM at the scale or funding level of the Chinese efforts. The instinct is the same on both sides: wrap a general model with education-specific corpora, value alignment, and safety layers.

And both systems are honestly uncertain about whether their 2030 targets are achievable. China’s goal of basic universal K-12 AI education by 2030 is binding in direction but not easy for a system with the world’s largest education scale. The U.S. has no comparable national target, which makes progress easier to claim and harder to measure.

What I Take From This

I am not arguing that the U.S. should copy China’s top-down model. American education governance is deliberately decentralized, and that decentralization produces real benefits: experimentation, local responsiveness, freedom from a single political vision of what AI education should be.

If I had to rank what actually moves the needle, in order of what is most achievable in the next two years, it would look like this:

A grade level curriculum core, built at the state level first. This is the most tractable move right now. States that already have AI guidance (California, and the 34-state cohort broadly) are the fastest path to converting guidance into standards, without waiting on federal action that has not historically produced binding curriculum.

Named, specific prohibitions rather than general encouragement. No AI-generated homework submissions, no unsupervised elementary use, explicit language on cognitive offloading. These are cheap to write into policy and give teachers cover they currently do not have.

Funded teacher capacity-building, not just access. The evidence from both countries suggests that access and mandate without funded time and tools does not produce depth. This is the one both countries are actually failing at, and it is the one that will not solve itself through policy language alone.

Rural and under-resourced-district funding, explicitly named as an AI-equity gap rather than folded into general E-Rate or Title I language.

The Chinese bet is that a binding curriculum plus prescriptive guardrails plus rural-equity funding plus education-domain AI tools will produce a generation of AI-literate citizens by 2030, and their own data suggests that bet is running into the same depth problem the U.S. has. The American bet is that state experimentation plus federal encouragement plus private-sector innovation will get there through emergence, with less top-down structure to fall back on when it does not. Both bets are honestly uncertain. Both systems could learn from the other, and neither has yet solved the part that matters most.

What would it look like if your state or district treated AI literacy as a binding curricular commitment, backed by funded teacher time rather than a mandate alone? That, I think, is the question worth sitting with this week.

References

AI for Education. (2025). State AI guidance for K12 schools. https://www.aiforeducation.io/ai-resources/state-ai-guidance

ChinaTalk / Ottinger, L. (2026, April 3). China’s AI education experiment. https://www.chinatalk.media/p/chinas-ai-education-experiment

Education Commission of the States. (2025). How states are responding to the rise of AI in education. https://www.ecs.org/artificial-intelligence-ai-education-task-forces

教育部基础教育教学指导委员会. (2025). 《中小学人工智能通识教育指南（2025年版）》；《中小学生成式人工智能使用指南（2025年版）》. Released May 12, 2025.

国家网信办, 国家发改委, 教育部, 等. (2023). 《生成式人工智能服务管理暂行办法》. Effective August 15, 2023. https://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm

国务院办公厅. (2017). 《新一代人工智能发展规划》. 国发〔2017〕35号.

罗生全, 谭爱丽, 钟奕军. (2023). 人工智能教育应用中的伦理风险及其规避. Education Sciences in China, 6(2).

U.S. Department of Education. (2025). Dear Colleague Letter on AI use in schools. https://www.ed.gov/about/news/press-release/us-department-of-education-issues-guidance-artificial-intelligence-use-schools

White House. (2025, April 23). Executive Order 14277: Advancing Artificial Intelligence Education for American Youth. https://www.federalregister.gov/documents/2025/04/28/2025-07368/advancing-artificial-intelligence-education-for-american-youth

王继新, 黄柳苍. (2025, April 17). 人工智能在教育应用中的伦理风险及防范. 中国社会科学网.