Three answers to the same question: should schools embrace AI?
Vietnam, China and New York City have taken very different approaches to artificial intelligence in schools this year. One is building a compulsory national curriculum, another is coordinating multiple agencies to bring AI and related tech into special education, and the third is moving to restrict student access to AI tools and limit in-class screen time. Those choices reflect different policy priorities, such as workforce preparation, inclusion and accessibility, or developmental caution, and they carry clear consequences for vendors, school systems and students.
What each jurisdiction announced
Vietnam: a fast national push
The Ministry of Education and Training announced an AI education framework on August 21. According to technode.global, the plan includes a pilot in academic year (AY) 2025-2026 and a compulsory rollout starting AY 2026-2027 that allocates 12 periods per academic year to AI lessons. The curriculum is sequenced by level: primary grades cover basic AI concepts and data protection; middle school adds digital-product development and AI principles; high school focuses on AI tool design and community-focused problem solving with an emphasis on “human‑centered thinking.” Technode.global also reported that Prime Minister Le Minh Hung has encouraged the rollout.
China: an interagency push for special education
On August 31, China’s Education Ministry and six other government departments issued a joint action plan to integrate AI, big data and virtual reality into special education, Xinhua reported. The plan stresses improving digital infrastructure to support accessibility and personalization. Xinhua cited system figures for 2025: 927, 000 students enrolled in special education schools, a 145% increase since 2012; 2, 464 special education schools; and 85, 100 special education teachers, in a country of “over 1.41 billion” people.
New York City: restriction and screen‑time limits
Documents obtained by Chalkbeat on September 2 show the New York City Department of Education (DOE) preparing a restrictive approach. The draft policy described a ban on student use of AI devices from 2‑K through eighth grade, classroom screen-time caps (30 minutes for elementary students and 45 minutes for middle school), and prohibitions on certain administrative and teacher AI uses such as grading, behavior monitoring, counseling and writing special-education plans. The DOE draft allowed limited teacher use for lesson planning, translation and communications only with board approval. Chalkbeat reported a final policy from the DOE was expected on September 9.
“Hands‑on learning, social interaction, play, and teacher‑guided instruction”, and “Routine individual screen use can crowd out those experiences, ” the DOE emphasized in documents obtained by Chalkbeat.
Why these differences matter
Policy is more than technology. It signals what societies value in schooling. Vietnam treats AI literacy as national infrastructure to teach across grades. China is targeting inclusion and personalization for students with disabilities through a coordinated, multi-department program. New York City’s draft prioritizes early-childhood development and protective limits on screen time.
Each path has benefits and risks. A rapid national curriculum can speed up basic AI literacy and create market opportunities for domestic edtech, but only if teacher training, hardware and connectivity accompany the rollout. Targeted tech in special education can widen access with adaptive tools, but it increases the need for data governance, clinical oversight and disability-informed procurement. Restrictive policies protect young children’s classroom time and social learning, but without clear exemptions they risk blocking assistive technologies some students depend on.
Common implementation pressure points
- Teacher capacity: Teachers need curriculum training, classroom coaching and assessment redesign, not just lesson plans.
- Infrastructure and equity: Devices, reliable maintenance and broadband funding determine whether programs reach rural and low-income students.
- Definitions and enforcement: Policymakers must be precise about what counts as an “AI device” versus an “AI-enabled service, ” and how IEP-based exemptions for assistive tech will work.
- Data governance: Student privacy, consent and data-minimization rules must be specified up front, including how vendors will comply with local laws.
- Evaluation: Pilots need pre-registered KPIs and public reporting on learning gains, equity and developmental impacts.
What this means for businesses and edtech vendors
These policy splits form your market map. Vietnam and China will likely open procurement windows for curriculum content, teacher professional development and assistive technologies. Vendors should prepare for strict requirements on transparency, privacy and explainability. In New York City and similar cautious districts, near-term opportunities will favor privacy-first assistive tools, teacher productivity aids that meet board approval, and pedagogically sound, low-screen alternatives.
Prioritized checklist for vendors (practical, actionable)
-
Immediate priorities (0-6 months)
- Assemble a procurement-ready documentation pack: Data Protection Impact Assessment (DPIA), model card, SOC 2 or equivalent security report, accessibility audit, privacy policy and sample parental consent workflows.
- Bundle teacher professional development: commit to at least 20 hours of teacher training and 10 hours of classroom coaching per 100 students in pilot proposals.
- Design low-bandwidth/offline modes: provide a version that functions under limited connectivity (target <256 kbps per concurrent user for critical features).
-
Medium-term priorities (6-18 months)
- Instrument explainability and audit logs: include human-readable rationales for model outputs, confidence scores and tamper-evident logs of data provenance, while balancing immutability with deletion and erasure rights.
- Build clinical and accessibility partnerships: involve special-education clinicians, therapists and disability advocates in product design and formal usability testing.
- Create an evaluation plan: predefine KPIs (see below), provide analytics dashboards for district partners, and set a three- to six-month pilot reporting cadence.
Vendors should also expect different contracting requirements across markets. National ministry tenders in Vietnam and China will focus on scale and standards compliance, while U.S. districts will emphasize FERPA and COPPA compliance, parental consent and local board approvals.
Procurement documentation checklist (quick reference)
- Data Protection Impact Assessment (DPIA) and privacy-by-design statement
- Model card explaining inputs, outputs, limitations and known biases
- Security attestation (SOC 2 or equivalent) and incident response plan
- Accessibility conformance report and clinical validation for special‑education features
- Teacher training curriculum and coaching plan with measurable hours
- Sample IEP integration and exemption workflows for assistive tech
How to run pilots that actually answer questions
Pilots should be treated as experiments with public accountability, not vendor demos. Recommended pilot design:
- Minimum duration: one academic year.
- Sample size: large enough to measure value-added learning gains, ideally multiple classrooms across diverse schools, urban, suburban and rural.
- Core KPIs: learning gains (value-added or pre/post measures), equity (disaggregated outcomes by income, region and disability), teacher adoption and confidence, and developmental metrics for younger students, such as attention and social interaction measured by validated scales.
- Transparency: pre-register KPIs and publish anonymized results and any privacy incidents.
Policy trade‑offs and technical caveats
Some technical proposals, like using distributed ledgers to log model decisions and data provenance, are floated as accountability solutions. They can provide tamper-evident records, but they also create trade-offs. Immutability can conflict with legal deletion rights, and distributed systems add complexity and cost that districts will scrutinize. Practical accountability usually starts with good logging, explainability, DPIAs and robust incident response plans, rather than a single silver-bullet technology.
Key questions readers are asking (and short, honest answers)
-
What exactly is Vietnam doing with AI in schools?
Vietnam’s Ministry of Education and Training announced an AI curriculum on August 21 that, according to technode.global reporting, includes a pilot in AY 2025-2026 and a compulsory rollout from AY 2026-2027 allocating 12 periods per academic year across primary to high school levels.
-
What is the focus of China’s August 31 action plan?
A joint action plan led by the Education Ministry and six other government departments aims to integrate AI, big data and VR into special education and to improve digital infrastructure for accessibility and personalization, according to Xinhua.
-
Is New York City banning AI in schools?
Draft documents obtained by Chalkbeat indicate the NYC DOE planned to prohibit student use of AI devices from 2‑K through eighth grade, set classroom screen‑time caps (30 minutes elementary, 45 minutes middle) and restrict certain administrative uses; Chalkbeat reported the DOE expected to issue a final policy on September 9.
-
Will special education get exceptions if AI is limited for other students?
China’s plan explicitly targets special education, while NYC’s draft suggests possible exemptions for assistive supports, but precise carve‑outs, definitions and enforcement rules need to be published for clarity.
-
What should school systems and vendors prioritize now?
Priorities are teacher professional development, clear data-governance contracts, rigorous pilot KPIs and explicit exemptions for assistive tech. Vendors should package PD, security and privacy documentation and low-bandwidth modes with their offers.
Parting practical thought
Decisions about AI in education are strategic choices that will shape who benefits and who is left behind. National curricula can jumpstart literacy and create market certainty. Inclusion plans can improve access for students with disabilities. Cautionary policies can protect early childhood development. The best path is neither blind acceleration nor blanket prohibition. Fund teachers and infrastructure, define precise exemptions, require public evaluation of pilots, and hold vendors to clear privacy and explainability standards. That is how education systems turn an emerging technology into durable learning gains rather than another uneven experiment.