HANOI, 31 AUG 2026 — Vietnam will teach artificial intelligence in every grade from one to twelve, starting with the 2026-27 school year, across 50,000 institutions and 27 million students. The compulsory core is 12 periods a year.
What the framework says
The Ministry of Education and Training issued the framework in mid-August, following a pilot in the 2025-26 year. It covers primary, secondary and high school, and the content spans human-centred thinking, AI ethics, AI techniques and applications, and AI system design.
The compulsory element is 12 periods per academic year. Beyond that, extension components are optional and each school decides for itself.
Twelve periods is a decision, not an oversight
A period is roughly 45 minutes, so the mandatory allocation is somewhere near eight or nine hours across a full school year. That is one lesson every three weeks.
As a curriculum, that is very little. As a policy instrument, it is a defensible design. The distinction matters, because most coverage will report the ambition and skip the dose.
A small compulsory core with a large optional extension is what a ministry writes when it wants universal minimum exposure without displacing existing subjects. Any larger allocation has to come out of mathematics, literature or science, and no education ministry makes that trade casually. Twelve periods buys every child the vocabulary and the ethics discussion, and leaves depth to schools that can staff it.
The practical consequence is that outcomes will diverge sharply. Schools with a capable teacher and equipment will build something on top of the core; schools without will deliver the core and stop. A framework designed this way produces a floor, not a standard, and the gap between them is wide.
The number that is missing
There are 1.4 million teachers in the ministry's database, and the framework as reported carries no teacher training provision and no budget figure.
That absence is the substantive issue and it is not a criticism unique to Vietnam. Curriculum reform fails at the teacher-supply step more reliably than at any other, because the ministry can mandate content by decree and cannot mandate the capacity to teach it. Twelve periods a year of AI ethics and system design, taught by someone with no preparation in either, becomes twelve periods of reading a textbook aloud.
A national AI competency framework is expected by June 2027, which is a year after teaching starts. The sequencing is telling: the standard for measuring competence arrives after the first cohort has already been taught.
Why the pilot year matters more than the announcement
The 2025-26 pilot contains the only actual evidence for how this will work, and its results have not been published.
The questions a pilot answers are exactly the ones that decide whether the rollout works: how many schools could deliver the core without additional staff, what fraction took up the optional extension, what equipment turned out to be necessary, and how the material landed with grade 1 pupils as against grade 12. None of those results appear in the announcement.
Nothing suspicious follows from that. Pilot data is often not published, and a ministry moving to national rollout has presumably read it. But an outside assessment of whether this will work has nothing to work with, and the confident reporting of a 27-million-student programme rests on a framework document rather than on a result.
Where this sits in Vietnam's wider push
This is the education leg of a programme that already has a legal leg and a diplomatic one.
We reported that Vietnam's AI law has been in force since March, with its deadlines starting next year, and that its technology statement with Singapore named six fields and no money. Prime Minister Le Minh Hung used the academic year review to press the education sector on digital transformation and AI in teaching.
The pattern is consistent. Vietnam is moving faster than its neighbours on declared AI policy, in governance and now in education, and slower on the resourcing that would make either bite. That is not hypocrisy; it is the normal shape of state capacity running behind state ambition, and it is visible in the same way in Malaysia's data-centre approvals and in Singapore's own workforce programmes.
What to watch instead of the headline
Three things would tell you whether this becomes real, and none of them is the curriculum document.
First, teacher training enrolment: that is the binding constraint, and a figure that will exist by mid-2027 whether it is published or not. Second, the uptake rate on the optional extension, which marks the difference between a floor and a real programme. And whether the June 2027 competency framework assesses students, teachers or both — assessing only students measures the intake, while assessing teachers measures the system.
For the region, the useful comparison is not whether Vietnam is ahead. It is that Vietnam has now published a number, 12 periods, which most neighbours have not, and a published number can be argued with.
