--- title: "China's schools get a GenAI rulebook: no copying homework, no exam data in chatbots" date: 2026-10-08 category: Policy & Governance site: NeuroAI canonical: https://neuroai.site/a/na-policy-school-genai-guideline language: en --- # China's schools get a GenAI rulebook: no copying homework, no exam data in chatbots > Two May 2025 ministry guidelines set tiered, grade-by-grade rules for generative AI in primary and secondary schools, with hard bans on exam-data input and AI-written homework. A 12-year-old pastes a math problem into a chatbot during class and copies the answer straight into her workbook. A teacher wonders whether the essay on his desk was written by a student or a model. Across China's classrooms, generative AI (生成式人工智能) arrived faster than any rulebook could, so the rulebook finally showed up. ## Two documents, one signal In May 2025, the Ministry of Education's Basic Education Teaching Guidance Committee released a paired set of documents: - the *Primary and Secondary School AI General-Education Guideline (2025 Edition)* (《中小学人工智能通识教育指南（2025年版）》), and - the *Primary and Secondary School Generative-AI Use Guideline (2025 Edition)* (《中小学生成式人工智能使用指南（2025年版）》). The first builds a *spiral curriculum (螺旋式课程)* — interest and basic awareness in primary school, technical principles in junior high, systematic thinking in senior high. The second is the governance half: it tells schools *how* the tools may and may not be used. State broadcaster CCTV and *People's Daily* both carried the release, marking it as a national baseline rather than a local experiment. ## The tiered-use ladder The guideline does not apply one rule to all ages. It sorts permissions by stage: - **Primary school (小学):** students must be helped by a teacher or parent when using open content-generation features, and are *prohibited from using them alone*. The goal is interest plus an *ethical introduction (启蒙)* to data privacy, not independent generation. - **Junior high (初中):** schools may let students moderately analyze the *logic* of generated text and cross-check its correctness, building critical habits. - **Senior high (高中):** students can run inquiry projects and are expected to assess the *social impact* of generated content themselves. The throughline is *graded autonomy with supervision* — the older and more capable the student, the more latitude, but never unattended. ## The hard bans Two prohibitions are stated without qualification: - **No AI-written homework or exam answers.** The guideline forbids students from directly copying generative-AI output as homework or test responses, and limits the abuse of AI in creative tasks, to stop "proxy" use at the source. - **No exam or identity data in the prompt.** Teachers and students are strictly forbidden from entering exam questions, personal identity information, or other sensitive data into generative-AI tools. To back this up, the document calls for a **dynamic whitelist (白名单)** system: a living list of generative-AI tools approved for campus use, set by local education authorities, with an ethical-review mechanism attached. ## Beyond the classroom wall The guideline also reaches the surrounding ecosystem: - **Evaluation:** schools are encouraged to explore folding *AI literacy* into the student *comprehensive quality evaluation (综合素质评价)*, building a growth file rather than a one-off test. - **Data security:** it mandates data-security management norms and privacy protection, explicitly covering the sensitive inputs schools must not leak. - **Multi-party support:** it asks government, schools, families, enterprises, and research institutes to share responsibility, so that AI education is "elastic, adjustable, and co-built" rather than imposed top-down. The framing is deliberately dual: *apply as king, governance as base (应用为王、治理为基)* — release the tool's teaching value, but keep a floor under safety and ethics. ## Why this matters outside China Few jurisdictions have issued grade-specific generative-AI classroom rules at this granularity. Most Western guidance stays at the school-district or university level. A national, age-tiered baseline from the world's largest K-12 system (China runs the largest education system on the planet) becomes a de-facto reference design that edtech vendors exporting to other markets will be measured against. For model providers, the whitelist and the exam-data ban create a concrete compliance surface: a product that cannot prove it blocks identity and test-content inputs will struggle to clear school procurement. ## What changes on a normal school day The guideline turns abstract worries into concrete classroom routines: - **Teachers become the gatekeepers.** Because primary students cannot use open generation alone, lesson plans must build supervised sessions rather than free periods with a chatbot. - **Verification becomes a skill, not a gesture.** Junior-high activities center on checking a model's logic and cross-referencing its claims, so assessment shifts from "did you write this" to "can you defend this." - **Senior projects get an impact lens.** High-school inquiry work is expected to weigh the social effect of generated content, nudging AI use from answer-machine toward research tool. This is deliberately conservative next to the unrestricted consumer chatbots students meet at home, and the gap is part of the point: school is where guardrails are taught before autonomy is granted. ## Honest limitations This article draws on the ministry committee's two guidelines as reported by CCTV, *People's Daily*, and China Education News; it does not assess how individual provinces or schools have implemented them, which will differ. The documents set principles and prohibitions but leave enforcement, auditing, and whitelist composition to local authorities we cannot verify here. No central per-student funding figure was attached to the guidelines, so cost impact is not quantifiable from the source. The whitelist mechanism's real effect depends on vendor cooperation we have not independently tested. ## What readers can do now - **If you build education AI, design for the bans by default** — block identity and exam-content inputs and log refusal, because the whitelist standard will expect it. - **If you are a parent or teacher, rehearse the "cross-check, don't copy" habit** with students, since the guideline rewards verification over reliance. - **If you study policy, watch the whitelist rollout in pilot provinces** as the first real signal of how strict campus AI access will become. --- Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-policy-school-genai-guideline Free to quote with attribution and a link to the original.