A government has AI deliver most teaching in a core subject

Where things stand (as of 11 Oct 2026)

I found no qualifying adoption since the question opened on 16 Sep 2026. The bar is high: the AI-led share must be official, cover the whole system, be compulsory, and exceed 50% of class time. The closest cases each miss at least one requirement.

• Israel, "English for Everyone" and Project 720. In June 2026 the Prime Minister's Office and the Education Ministry announced that all middle schools would move to AI-driven personalised English learning from 2026-27. That is before the window opens, and no share of class time was stated.
  • The ministry calls it "AI-assisted" and "teachers first". Teachers are described as mentors, and the app only goes live in December (Times of Israel, 21 Jun 2026 (archived copy); Israel Hayom, 19 Aug and 28 Sep 2026).
  • The vendor request for information mentions an "independent and full solution for English learning". It is still only procurement, which the question excludes.
  • Project 720 itself is a voluntary pilot that schools apply to join (call for participants; JNS, Oct 2026).
  • The question's own exclusion of labels like "personalized learning" and "teacher as facilitator" fits this case. Israel votes on 27 Oct 2026, which adds policy uncertainty.
• El Salvador. Grok-based AI tutors are going to all public schools within about 18 months, but as support for teachers. The minister said "Jamás vamos a sacar a los docentes" ("we will never remove teachers"). The World Bank-funded programme is built around teachers (Infobae, 5 Sep 2026; El Pregonero).
• Brazilian states. These are the nearest on paper.
  • Paraná puts 1 of its 2 weekly English lessons on an adaptive EF platform, exactly 50% (UEM master's thesis, 2026).
  • São Paulo requires only a minimum number of lessons on its SPeak platform (Seduc-SP bulletin, March 2025).
  • In São Paulo's 2026 night-school expansion, final-year students take English only through SPeak. That is a subset of the grade, not the whole cohort, and it predates the window.
  • Brazil's National Education Council approved national AI guidelines on 1 Sep 2026: AI must not replace teachers' "pedagogical mediation" (G1).
• Momentum is against screens.
  • The New York City and Los Angeles districts have banned or restricted student AI use.
  • In Texas, the State Board of Education rejected Alpha School's AI charter. Districts using Alpha's tool say they use it only as a supplement (Texas Tribune, 1 Oct 2026).
  • The UAE is regulating digital tools in Grade 6 after its own study favoured non-digital classrooms (Khaleej Times, 27 Aug 2026).
  • New South Wales will cap classroom screen time from 2028.
  • China's "AI Plus Education" plan uses AI to support teachers (Xinhua, Apr 2026), and OpenAI's country programmes are framed as "teacher-first".

Base rate

In more than ten years of adaptive learning platforms and four years since ChatGPT, no government has met these criteria; the closest are 50% and subset-only cases. The only historical parallel is educational television in the 1960s–70s (American Samoa, El Salvador, Côte d'Ivoire). A few governments then made broadcast lessons central to instruction across their systems, though probably not above 50% of class time. That suggests a hype cycle can produce such adoptions occasionally, but rarely.

Paths to YES and NO

YES most plausibly comes from one of these:

• a Brazilian state putting a whole grade's English mostly on an AI platform in its official timetable;
• Israel's ministry setting a majority-AI English model in a formal circular;
• a short-staffed system (Vietnam's compulsory English, Gulf states, African or Indian states) assigning a subject to AI tutors;
• a much faster rise in AI capability around 2028–2031.

Hundreds of eligible governments exist, and only one has to qualify.

NO is driven by:

• the strict documentation test;
• teachers' unions and parents;
• the screen backlash;
• the political cost of saying AI replaces teachers;
• slow scaling from pilot to whole system.

Model

I used rising quarterly hazards: 0.15% in Q4 2026, about 0.2–0.35% per quarter in 2027, and about 1% per quarter in 2031. The cumulative risk is about 1.2% by end-2027, 5.7% by end-2029 and 12.6% by end-2031, computed in Python as one minus the running product of (1 − hazard).

Strongest case against my estimate: if AI reaches roughly human-teacher quality by 2029, a short-staffed government could mandate AI-led language or maths teaching quickly, which would push the result above 20%. I weight that partly. The opposite case is that the explicit-share test alone keeps the answer near 5%.

Question. By the last day of each quarter, has any national, state, or provincial government adopted a policy under which AI systems, not teachers, deliver most instruction in a core subject across its state schools?

Definitions.

• Government: a sovereign national government or a first-order constituent state or province, including an equivalent regional government with education-policy authority. Municipalities, school districts, individual schools, and private school networks do not qualify.
• State schools: publicly operated primary or secondary schools, together with publicly funded charter or equivalent schools where they form part of the jurisdiction's public-school system. Privately operated schools outside that system and tertiary education are excluded.
• Across its state schools: the policy establishes a system-wide instructional model for at least one complete grade/year cohort in all ordinary state schools offering that grade and subject within the adopting government's jurisdiction. Exceptions for special education, individual accommodations, or emergency closures are allowed; selection of particular districts, schools, volunteers, or experimental cohorts is not. A national policy applying only to a small network of directly administered schools does not qualify.
• Core subject: mathematics, the main language of instruction/literacy, a compulsory additional language, natural science, or history/social studies, taught as a required curricular subject. An elective, extracurricular program, or course solely about AI use does not qualify.
• AI systems deliver most instruction: more than 50% of the subject's scheduled instructional time over a complete academic year is assigned to direct AI-led teaching, including explanations, guided practice, and individualized instructional feedback. The system must generate or adapt instruction in response to students; static videos, ordinary digital textbooks, and fixed answer-marking software are insufficient. Human supervision, safeguarding, motivation, curriculum oversight, and supplementary teaching are compatible with qualification. Time in which a teacher delivers a lesson using AI-generated materials or suggestions is human-led, not AI-led. Homework and optional tutoring outside scheduled instruction are excluded from the calculation.
• Adopt: enact a law, issue a final regulation, or formally approve a ministry/education-department policy establishing this instructional model as the required or default provision, with a stated implementation date or timetable. Classroom implementation need not already have begun. Merely permitting schools to opt in does not qualify.

Resolves YES if an adopted policy and its official supporting documents establish all the definitions above. The documents must specify an AI-led share exceeding 50%, a timetable from which that share can be calculated, or explicitly assign the majority of scheduled subject instruction to AI rather than human teachers. A phased rollout counts upon adoption only if its approved end state meets the system-wide requirement without requiring a further policy decision.

Resolves NO / Exclusions. No qualifying adoption has been documented by the quarter-end. Draft legislation, consultation documents, aspirations, procurement alone, optional tools, isolated charter approvals, pilots without an adopted system-wide rollout, and supplementary AI tutoring do not qualify. Labels such as “AI-powered schools,” “personalized learning,” or “teacher as facilitator” alone do not establish majority AI-led instruction. Nationwide provision of AI textbooks does not count unless the instructional-time and teacher-substitution requirements are met.

Resolving source. The adopting government's official gazette or legislative register for enacted laws; its official regulatory register for regulations; or final policy documents and implementation directives published by its ministry or department of education. Official annexes, curriculum timetables, and implementation guidance may establish scope and instructional shares. Vendor statements, press reports, and school publicity cannot independently resolve the question. These are the narrower resolving sources for this question; publication dating follows Resolution date (A3), and the Resolution window applies. A newly published account of an adoption before the opening date does not count as a new adoption.