# EduPolicyAtlas global generative-AI school policy research expansion

**Information checked through: September 8, 2026. Overall confidence: high for the ten audited atlas records and the strongest new profiles; moderate for the broader global inventory; low to unknown where expressly marked as a research lead.**

## Executive research brief and scope

EduPolicyAtlas should **not** evolve into a map of where “AI is allowed” or “AI is banned.” The strongest evidence from the current atlas and the expanded research shows that the practical answer usually depends on at least four independent questions: whether a rule is binding, the learner's age and supervision, whether the activity is assessed, and whether the tool or data flow has been approved. France, Singapore, New Zealand, England, and New York City demonstrate how misleading a one-dimensional label can be. citeturn4view3turn9view1turn9view0turn12view0turn14view1

I inspected the live EduPolicyAtlas and its HTML research-report route. The site currently presents ten jurisdiction starters and explicitly describes itself as a starting point rather than globally representative coverage. citeturn1view0turn2view0turn3view0 The raw `/research-report.md` resource could not be fetched through the available browser because of its content type, so I did **not** treat it as reviewed. No policy or news export files were attached to this conversation, and I did not locate a publicly exposed export endpoint. Accordingly, any differences below are independent verification or proposed additions rather than recovery of unpublished source mappings.

This pass supports roughly **thirty substantive jurisdiction or education-system records** at varying depths, including the original ten, plus a larger discovery queue. It covers every requested world region, but it is **not comprehensive global coverage**. The strongest expansion opportunities are Norway, Denmark, Italy, Taiwan, Indonesia, Malaysia, Kazakhstan, British Columbia, Quebec, Wales, Argentina, Saudi Arabia, Jamaica, Trinidad and Tobago, and several Pacific systems. Some of these have dedicated school-AI guidance; others deliberately belong in “general rules / implementation / policy under development” categories.

The biggest findings are:

| Finding | Atlas consequence | Confidence |
|---|---|---|
| New York City now has a true **student-facing GenAI moratorium for grades 2K-8**, not merely a vague “restriction,” with accessibility exceptions and separate permitted high-school pilots. AI also may not make grading, placement, promotion, graduation, or behavioral decisions. | Immediate correction and expanded record. | **High** citeturn14view1turn14view2turn14view3 |
| France's June 2025 framework is much more operational than a simple age-banded summary suggests. Primary pupils do not directly manipulate GenAI; use begins on a limited, teacher-supervised basis from 4e; lycée students can use it more independently inside teacher-defined boundaries. Unauthorized substitution of AI for the student's work can be fraud, and AI detectors are discouraged as unreliable. | Expand age, assessment, accounts, detectors, and privacy fields. | **High** citeturn4view3turn6view0turn6view1turn6view2 |
| Singapore has unusually precise age and assessment distinctions. Primary 1-3 pupils are not assigned work requiring direct AI; Primary 4-6 use education-specific AI under supervision; national examinations prohibit AI, while some coursework can allow it under defined conditions. | Upgrade from a broad “staged” description to rule-level data. | **High** citeturn9view1 |
| New Zealand's assessment position is stricter and more concrete than the current short summary suggests. Schools assessing relevant standards need authenticity arrangements addressing GenAI, evidence submitted for assessment must be the student's own, and GenAI is prohibited in NCEA external assessment. | Add examination/authenticity fields. | **High** citeturn9view0 |
| England remains locally discretionary for classroom adoption, but that does not mean “anything goes.” Statutory safeguarding, data-protection, age, filtering/monitoring, IP and assessment obligations constrain local choice. Product-safety expectations were also updated in January 2026. | Model local discretion separately from legal constraints and procurement safeguards. | **High** citeturn12view0turn12view1 |
| Norway moved toward one of Europe's clearest age-based national positions in June 2026: grades 1-7 generally should not be given AI access, grades 8-10 may begin cautious use after teacher preparation, and upper-secondary pupils should learn appropriate use. | High-value new record, but distinguish the government instruction/recommendation process from statutory prohibition. | **High for announcement; moderate for final implementation text** |
| Kazakhstan is moving beyond AI-literacy curriculum into explicit national governance for secondary education. Its 2026 policy program addresses assessment, academic integrity, personal data, teacher responsibility and AI-generated content. National Academy materials also provide detailed model rules. | High-value Central Asian addition, but separate national framework from locally adopted model rules. | **Moderate-high** |
| In much of Africa and the Pacific, the evidence located concerns national AI strategy, digital education, curriculum, pilots or teacher capacity rather than a dedicated classroom-GenAI permission code. | Publish “dedicated policy not located” or “general rules apply,” never “no policy.” | **Moderate** |
| Rigorous K-12 outcome evidence remains much thinner than the policy activity. England explicitly notes limited evidence on learner development and safety, while Ireland describes its guidance as an initial, regularly reviewed step in an early evidence base. | Avoid “best policy” rankings and causal claims. | **High** citeturn12view0turn9view3 |

For the master geographic inventory, use the UN Statistics Division M49 country-and-area reference as the neutral backbone: `https://unstats.un.org/unsd/methodology/m49/`. Territories should receive their own atlas jurisdiction only when they have a distinct education authority or materially distinct school rules. For disputed jurisdictions, record the issuing authority and territorial scope without using the atlas to adjudicate sovereignty.

## Coverage ledger and existing-record audit

**Research-status vocabulary**

| Status | Definition |
|---|---|
| **Dedicated policy found** | An authoritative education-specific document materially governs AI/GenAI use. |
| **Existing general rules apply** | Authoritative privacy, assessment, education, procurement, AI or digital rules clearly constrain school AI, but no single dedicated classroom policy controls the question. |
| **Explicitly verified absence** | An authoritative body expressly says a dedicated policy does not exist. This label should be rare. |
| **Not located** | The searches performed did not locate a dedicated policy. This is not evidence of absence. |
| **Not researched sufficiently** | Search depth was insufficient to support even a “not located” conclusion. |
| **Policy under development / pilot** | An authority has announced development, consultation or implementation without a complete operative rule set. |

**Coverage and search ledger**

| Jurisdiction / system | Level | Languages searched or sources reviewed | Result | Depth | Main unresolved issue |
|---|---|---|---|---|---|
| Australia | National framework | English | Dedicated national framework | Deep | State/territory overlays still need systematic extraction |
| Japan | National | Japanese, English translation | Dedicated GenAI guideline v2.0 | Deep on authority/version, moderate on rules | Full Japanese rule-level extraction |
| France | National | French | Dedicated framework | Deep | Local implementation examples |
| England | National | English | National guidance + product/assessment safeguards | Deep | School/local policies vary |
| New Zealand | National | English | Dedicated national guidance + assessment rules | Deep | Tool-specific age terms remain volatile |
| Singapore | National | English | Detailed staged national approach | Deep | Pilot-specific local implementation |
| Chile | National | Spanish | National guidance initiative | Moderate | Full rule-by-rule extraction |
| Ireland | National | English | Dedicated living guidance | Deep | Separate examination rules need further granularity |
| Washington State | State | English | State human-centered AI guidance | Moderate-high | District implementation and version lineage |
| New York City | District | English | Binding/local operating policy for 2026-27 | Deep | Pilot evaluation data |
| British Columbia | Province | English | Provincial AI/digital-literacy guidance | Moderate-high | Assessment and district overlays |
| Quebec | Province | French | Provincial pedagogical/ethical/legal GenAI guidance | Moderate-high | School-service-center implementation |
| Wales | National education system | English, Welsh environment | Hwb guidance; inspectorate implementation evidence | Moderate-high | No single national permission code |
| Argentina | National | Spanish | PaideIA implementation program and GenAI guidance | Moderate | Detailed assessment rules |
| Saudi Arabia | National | Arabic | Ministry/SDAIA school GenAI guide located | Moderate | Page-level extraction still needed |
| Taiwan | National | Traditional Chinese | Primary/secondary GenAI guidance, assessment/homework guidance | High-moderate | Confirm current v2.1 wording against all predecessor notices |
| Malaysia | National | Malay | 2026 AI Literacy Guide located | Moderate | Page-image/legal-authority review before public rule quotation |
| Indonesia | National | Indonesian | AI curriculum guidance plus 2026 cross-ministry education AI framework | Moderate-high | Primary-text verification of reported red/yellow/green assessment model |
| Norway | National | Norwegian | 2026 age-banded national recommendation announced | High-moderate | Confirm final Udir version and commencement wording |
| Denmark | National | Danish | Folkeskole GenAI recommendations + exam prohibition | High-moderate | Municipality implementation |
| Italy | National | Italian, official English rendering | National AI-in-schools guidelines | High-moderate | More classroom-level permission examples |
| Netherlands | National | Dutch | AI Act implementation/procurement framework; national digital plan developing | Moderate-high | Pending national plan and absence of a single classroom-use code |
| Kazakhstan | National | Kazakh, Russian, English govt rendering | National AI-in-education framework plus detailed Academy model rules | High-moderate | Legal status of each Academy rule and Sep. 2026 implementation instruments |
| Uzbekistan | National | Uzbek, Russian/English legal database | AI ethics law + AI curriculum mandate | Moderate-high | Dedicated school GenAI classroom code not located |
| Jamaica | National | English | Draft school-AI guidelines + national AI recommendations + pilots | Moderate | Draft guidelines not yet verified as final |
| Trinidad and Tobago | National | English | AI-supported rollout, strategic plan, AI-in-education policy planned | Moderate | National AI-in-Education Policy still under development |
| Fiji | National | English | General education/digital framework, AI integration activity | Moderate | Dedicated GenAI classroom rule not located |
| Tonga | National | Tongan/English policy environment | Education framework, general technology duties | Moderate | Dedicated GenAI policy not located |
| Samoa | National | English | ICT-in-education policy located, but 2018-23 term expired | Moderate | Current replacement status |
| Papua New Guinea | National | English | 2026 sovereign AI/government framework | Moderate | K-12 classroom-specific rules not located |
| Rwanda | National | English | Education AI pilots/working activity | Discovery+ | Dedicated school-use policy not established |
| Nigeria | National | English | AI/digital curriculum developments | Discovery+ | Classroom GenAI conditions |
| Hong Kong | Education system | Chinese, English | AI literacy, training and implementation sources | Discovery+ | Dedicated pupil-use code not established in this pass |
| India / CBSE | National board | English | AI curriculum evidence | Discovery+ | Classroom GenAI and assessment rules require separate board-level audit |
| South Africa | National/provincial | English | National AI-policy activity located | Limited | Dedicated school policy not located in completed pass |
| Kenya | National | English | Search initiated | Insufficient | Full ministry and KNEC audit |
| Pakistan | National/provincial | English/Urdu needed | Not sufficiently researched | Insufficient | Federal/provincial split |
| Bangladesh | National | English/Bangla needed | Not sufficiently researched | Insufficient | Ministry/assessment rules |
| Nepal | National | English/Nepali needed | Not sufficiently researched | Insufficient | Curriculum/exam sources |
| Sri Lanka | National | English/Sinhala/Tamil environment | General AI/digital statements located | Limited | School-use rules |
| Philippines | National | English/Filipino | Not sufficiently researched | Insufficient | DepEd policy and assessment rules |
| Thailand | National | Thai needed | Not sufficiently researched | Insufficient | OBEC/ministry sources |
| Vietnam | National | Vietnamese needed | Not sufficiently researched | Insufficient | Ministry guidance |
| Cambodia | National | Khmer needed | Not sufficiently researched | Insufficient | Ministry guidance |
| Solomon Islands | National | English | Policy-development reporting located | Limited | Primary ministry text |
| Vanuatu | National | English/French/Bislama environment | Older ICT framework leads | Limited | Current AI-specific policy |
| Cook Islands | National education system | English | Current ICT policy located | Moderate | Dedicated AI policy not located |
| Tajikistan | National | Tajik/Russian needed | AI education pilots discovered | Limited | Dedicated school-use rules |
| Kyrgyzstan | National | Kyrgyz/Russian needed | AI strategy activity discovered | Limited | Education-specific rules |
| Turkmenistan | National | Turkmen/Russian needed | Education/AI pilots discovered | Limited | Public policy documentation |
| Barbados | National | English | Regional CXC relevance identified | Insufficient | Ministry-specific AI rules |
| Guyana | National | English | Regional AI governance lead | Insufficient | School and CXC implementation |
| Dominican Republic | National | Spanish | National AI strategy lead | Insufficient | School GenAI-specific rules |
| Bahamas | National | English | Digital-governance lead | Insufficient | School AI rules |

The ledger therefore makes a crucial distinction: a place such as Uzbekistan can have binding nationwide AI ethics rules and mandated school AI instruction without yet having a clearly located GenAI homework policy; Wales can have privacy and implementation guidance without a single national “permission table”; Fiji can actively encourage AI integration while still lacking a located detailed pupil-use code.

**Audit of the current ten atlas records**

| Existing atlas statement / shorthand | Finding | Proposed replacement | Source | Reason / remaining uncertainty |
|---|---|---|---|---|
| **Australia: framework-led** | Correct, but underspecified. The Australian Framework explicitly addresses assessment-design instructions, attribution, human accountability, accessibility, privacy, retention, data sharing and cybersecurity. The national page was modified June 17, 2025 after ministers endorsed the 2024 Framework Review. | “National, principles-based guidance. Schools should make permitted/prohibited GenAI use explicit in assessment design, maintain human accountability, protect student data, support accessibility, and apply academic-integrity and attribution expectations. State and territory implementation may add requirements.” | `https://www.education.gov.au/schooling/resources/australian-framework-generative-artificial-intelligence-ai-schools` citeturn4view1turn5view1turn7view0turn7view1 | Distinguish 2025 review endorsement from a wholly new national instrument. **High confidence.** |
| **Japan: scenario-based guidance** | Correct but versioning matters. MEXT replaced/revised the July 2023 provisional approach with **Ver.2.0 on Dec. 26, 2024**. MEXT provides Japanese original and a tentative English translation. | “MEXT’s current national reference point is *初等中等教育段階における生成AIの利活用に関するガイドライン（Ver.2.0）*, Dec. 26, 2024. Treat it as guidance, not a simple national permission or ban; retain Japanese as controlling source and label the ministry’s English version as tentative.” | `https://www.mext.go.jp/a_menu/other/mext_02412.html` citeturn4view2 | Full Japanese rule-level extraction remains unfinished. **High confidence on version/status, moderate on unextracted dimensions.** |
| **France: age-banded** | Correct, but the actual framework resolves several “local detail” fields. Primary learners learn about AI without directly manipulating GenAI. From 4e, limited teacher-supervised use is contemplated. Lycée use can be more autonomous inside teacher-defined boundaries. Students should not have to create personal accounts; confidential/personal data must not be submitted. Unauthorized substitution can constitute fraud. AI detectors are discouraged. | Replace with a multi-field record rather than “age-banded.” Include age stage, supervision, account rule, personal-data rule, assessment fraud definition, AI-detector stance and required AI literacy. | `https://www.education.gouv.fr/cadre-d-usage-de-l-ia-en-education-450647` citeturn4view3turn6view0turn6view1turn6view2turn7view3 | One of the highest-value immediate upgrades. **High confidence.** |
| **England: local discretion** | Directionally right but too permissive sounding. Settings choose uses, but safeguarding, filtering/monitoring, vendor ages, data protection, IP and qualification rules constrain those choices. Pupil use should have appropriate safeguards and close supervision. | “Local adoption within national statutory and regulatory guardrails. Schools may choose use cases, but must address safeguarding, filtering and monitoring, age restrictions, data protection, IP and formal-assessment rules. Pupil-facing use requires particular care and supervision.” | `https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education` citeturn12view0 | Add Jan. 2026 product-safety expectations as complementary source, not replacement. citeturn12view1 **High confidence.** |
| **New Zealand: assessment safeguards** | Correct but incomplete. Schools with relevant consent-to-assess responsibilities need authenticity arrangements addressing GenAI. Assessed evidence must be the learner's own; GenAI is prohibited in NCEA external assessment. Ministry guidance also discourages personal-data submission and leaves operational AI policy to schools. | “School-managed classroom use under privacy and authenticity safeguards. AI-generated evidence cannot substitute for the learner's own assessed work; NCEA external assessments prohibit GenAI. Schools should maintain practical AI-use and authenticity policies.” | `https://www.education.govt.nz/school/digital-technology/generative-ai` citeturn9view0 | Vendor age/parental-consent terms should be timestamped because they change. **High confidence.** |
| **Singapore: staged, age-banded** | Correct and capable of much more specificity. P1-3 should not receive assignments requiring direct AI; P4-6 can use education-specific AI under supervision; some secondary work can use AI with disclosure/citation; national examinations prohibit it. | “P1-3: awareness without assigned direct AI use. P4-6: supervised education-specific tools. Secondary: conditional task-level use, including disclosure where required. National examinations: AI prohibited. MOE platforms and commercial tools are subject to data safeguards.” | `https://www.moe.gov.sg/news/parliamentary-replies/20260506-ai-usage-in-schools` citeturn9view1 | Add assessment and data fields. **High confidence.** |
| **Chile: guidance-led** | Supported. The Ministry's *PotencIA el aprendizaje* environment is a national education initiative/resource base rather than evidence of a single binding permission code. | “National guidance and capacity-building approach. Do not infer blanket pupil permission from the existence of the initiative. Add specific classroom/assessment claims only after document-level verification.” | `https://ciudadaniadigital.mineduc.cl/ia/` citeturn9view2 | Detailed rule extraction remains incomplete. **Moderate confidence.** |
| **Ireland: living guidance** | Correct. October 2025 guidance says human educators remain the final checkpoint, personal/sensitive data should be avoided, platform age/consent requirements must be checked, and existing school policies should be updated. A separate AI policy is not required. | “Living national guidance for responsible school AI use. Human review remains central; avoid personal or sensitive data; check vendor age/consent terms; update relevant existing policies rather than assuming a separate AI policy is mandatory; apply separate State Examinations Commission rules to assessed work.” | `https://assets.gov.ie/static/documents/dee23cad/Guidance_on_Artificial_Intelligence_in_Schools_2025.pdf` citeturn9view3turn10view0turn10view1turn11view0 | **High confidence.** |
| **Washington State: local discretion / foundations guidance** | Substantively reasonable. OSPI's January 2024 guidance has a version lineage that includes later 2024 updates/editions, so a single “July 1, 2024 publication date” obscures document history. | “Washington OSPI provides statewide human-centered AI guidance, while district implementation remains important. Store original publication, edition/revision and retrieval dates separately.” | `https://ospi.k12.wa.us/student-success/resources-subject-area/human-centered-artificial-intelligence-schools` citeturn13search0 | Exact current-edition document metadata should receive another document-level check before editing archival dates. **Moderate-high confidence.** |
| **NYC: restricted through grade 8** | Too vague. For 2026-27 NYCPS describes a **moratorium/banned student-facing GenAI use in grades 2K-8**, with exceptions such as IEP/504 assistive technology and continued ELL supports. High school adds mandatory AI-literacy modules and controlled approved pilots. AI cannot be used for grading or consequential student decisions. | “2026-27: student-facing generative AI is under a moratorium in grades 2K-8, subject to identified accessibility/support exceptions. High-school students receive AI literacy and may encounter approved supervised pilots. Staff use requires approved/privacy-reviewed tools. AI may not determine grades, behavior, placement, promotion, graduation, or similar student decisions.” | `https://www.schools.nyc.gov/about-us/policies/guidance-on-artificial-intelligence` citeturn14view1turn14view2turn14view3 | **Immediate correction. High confidence.** |

## Expanded jurisdiction profiles and cross-jurisdiction comparisons

The best publication candidates are those where the authority, current document, scope and operational consequences can already be separated clearly.

**Norway. Original source language: Norwegian. Interpretation based on the original government page.** On June 19, 2026, the government announced age-adapted national recommendations for grunnopplæringen. Grades 1-7 should generally not be given access to AI for schoolwork. Grades 8-10 may begin gradual, cautious use once teachers have sufficient competence. Upper-secondary pupils should learn to use AI appropriately for further study and employment. The government expressly contemplated exceptions where AI-based tools support needs such as language learning or individualized education. This is an important example of why “ban” is too crude: the position is age-banded, pedagogically conditional and exception-sensitive.

Source: `https://www.regjeringen.no/no/aktuelt/kunstig-intelligens-skal-i-all-hovedsak-ikke-brukes-i-barneskolen/id3166807/`

**Denmark. Original source language: Danish. Interpretation based on the original ministry text extraction.** The Ministry's *Anbefalinger om brug af generativ AI i undervisningen i folkeskolen* expressly says there are not nationwide classroom rules determining exactly how GenAI may be used in teaching. Under the Folkeskole framework, school boards approve teaching materials and local schools/teachers make instructional choices. This local discretion ends at national examination boundaries: pupils may not use GenAI in written or oral Folkeskole examinations, including preparation that contributes to the assessment basis. School leadership/municipalities must also ensure GDPR compliance.

Source: `https://uvm.dk/media/dylly3ye/250626-tilgaengeliggjort-anbefalinger-om-brug-af-generativ-ai-i-undervisningen-i-folkeskolen.pdf`

**Italy. Original source language: Italian; an English rendering is also published by the ministry.** The Ministry of Education and Merit adopted *Linee guida per l’introduzione dell’Intelligenza Artificiale nelle Istituzioni scolastiche* in 2025. The framework addresses principals, administrators, teachers and students across school grades, with emphasis on governed institutional adoption rather than consumer-chatbot permission. Schools acting as data controllers must examine necessity and proportionality, transparency and vendor arrangements; the guidance calls for a DPIA before relevant AI personal-data processing and links high-risk use to AI Act fundamental-rights assessment requirements. It is therefore especially useful for EduPolicyAtlas's procurement/privacy dimension.

Sources:  
`https://www.mim.gov.it/documents/20182/0/MIM_Linee+guida+IA+nella+Scuola_09_08_2025-signed.pdf/b70fdc45-4b75-1f7e-73bf-eab12989b928?t=1756468797694&version=1.0`  
`https://www.mim.gov.it/documents/20182/0/m_pi.AOOGABMI.Registro+Decreti%28R%29.0000166.09-08-2025.pdf/1062ccd9-d807-060e-5aa8-368e75980744?t=1756745377675&version=1.0`

**Taiwan. Original source language: Traditional Chinese.** The Ministry's current portal lists *中小學使用生成式人工智慧注意事項2.1*, rendered here as “Precautions for the Use of Generative Artificial Intelligence in Primary and Secondary Schools v2.1.” A separate ministry line of guidance addresses GenAI in student learning assessment and homework. Available ministry-linked notices stress teacher/parent guidance, service-age and account requirements, source verification, privacy, acknowledgment and avoiding overdependence. The coexistence of general-use guidance and assessment/homework guidance makes Taiwan a strong model for the atlas document-to-rule architecture. citeturn28search0turn28search12turn28search14turn28search16

Portal: `https://pads.moe.edu.tw/download.php`

**Indonesia. Original source language: Indonesian.** Ministerial Decision No. 127/P/2025 established implementation guidance for coding and AI from early-childhood through secondary education beginning in the 2025-26 school year, including curriculum and teacher-capacity arrangements. In March 2026 a seven-ministry joint decision created a broader framework for AI and digital technology across formal, non-formal and informal education, with child protection, human-centered adoption, privacy, family involvement and monitoring. A reported red/yellow/green model for AI in assessment is highly promising but should **not yet be published as a primary-source-verified rule** until the exact clauses are checked directly in the joint decision.

Sources:  
`https://smk.kemendikdasmen.go.id/produk-hukum/keputusan-menteri-pendidikan-dasar-dan-menengah-republik-indonesia-nomor-127p2025-tentang-pedoman-implementasi-koding-dan-kecerdasan-artifisial-pada-pendidikan-anak-usia-dini-jenjang-pendidikan-dasar-dan-jenjang-pendidikan-menengah`  
`https://www.menpan.go.id/site/berita-terkini/berita-daerah/pemerintah-terbitkan-skb-tujuh-menteri-tentang-pemanfaatan-ai-dan-teknologi-digital-di-dunia-pendidikan`

**Malaysia. Original source language: Malay.** The Ministry of Education's 2026 *Buku Panduan Literasi Kecerdasan Buatan (AI)*, “Artificial Intelligence Literacy Guide,” addresses institutions under the ministry and covers AI governance, teaching and learning, ethics, privacy/data security, academic honesty, bias and cyber safety. It is a stronger national source than generic “Malaysia is adopting AI” reporting, but the PDF still needs page-image verification before the atlas publishes fine-grained quotations or exact locators.

Source: `https://www.moe.gov.my/storage/files/shares/Pengumuman/BUKU%20PANDUAN%20LITERASI%20KECERDASAN%20BUATAN%20AI%209.6.26.pdf`

**Kazakhstan. Original source languages: Kazakh and Russian; an English government rendering exists for the presidential program.** Kazakhstan is one of the most significant discoveries in this pass. The President ordered a 2026-29 plan for AI in secondary education, including personalized learning, infrastructure, teacher preparation and protection of student personal data, with standards covering educational-content generation, assessment and academic integrity. The Ministry subsequently described a national conceptual framework for AI in secondary and vocational education. The National Academy of Education has also issued model rules defining GenAI, approved-tool “white lists,” acceptable and prohibited uses, special treatment for children under 13 or a service's applicable age, personal-data restrictions, assessment controls and procedural protections around AI-detector evidence. The Academy document says local rules take effect when approved by the head of the educational organization, so **its individual provisions should not be mislabeled as automatically binding national law for every pupil**.

Primary sources:  
`https://www.akorda.kz/ru/o-merah-po-vnedreniyu-iskusstvennogo-intellekta-v-sistemu-srednego-obrazovaniya-respubliki-kazahstan-1242737`  
`https://www.gov.kz/memleket/entities/edu/press/news/details/1070018?lang=kk`  
`https://uba.edu.kz/storage/app/media/22%20FD%20FD%20RS%20RS.pdf`

**Uzbekistan. Original source language: Uzbek, with Russian and English legal-database renderings.** Presidential Decree PF-189 of October 22, 2025 requires AI topics in general-secondary curricula from 2026-27 and establishes an “AI Day” program, textbooks/materials and teacher preparation. Separately, 2026 AI ethics rules impose broader principles of legality, human oversight, transparency, privacy, nondiscrimination and final human responsibility. These are material overlapping rules, but the research did not locate a dedicated national GenAI-homework/exam permission code. The correct atlas classification is therefore **existing national AI rules and curriculum mandates apply; dedicated classroom GenAI policy not located**, not “no policy.”

Sources:  
`https://www.lex.uz/en/docs/7790236`  
`https://www.lex.uz/en/docs/8083233`

**Saudi Arabia. Original source language: Arabic.** The Ministry of Education and Saudi Data and AI Authority issued a guide for generative AI in general education in January 2025, explicitly addressing students, teachers and parents and responsible/ethical use. This is a strong Middle East addition, but a second pass should extract exact age, assessment and data provisions page by page before public-facing rule claims go beyond that verified scope.

Source: `https://www.moe.gov.sa/ar/mediacenter/MOEnews/Pages/news1_24012025.aspx` citeturn22search1

**Argentina. Original source language: Spanish.** The national PaideIA program was launched May 12, 2025 to integrate AI across primary and secondary education. Its principles include computational thinking in primary education, critical and ethical AI use across primary/secondary levels, and more advanced AI development at secondary level. The ministry explicitly frames AI as support rather than replacement for teachers and rejects treating AI as infallible or neutral. This is best classified as a nationally led curricular and pedagogical implementation program, not a simple permission code.

Source: `https://www.argentina.gob.ar/noticias/capital-humano-lanza-el-programa-paideia-para-integrar-inteligencia-artificial-en-el` citeturn21view1

**British Columbia.** The province's “Digital literacy and the use of AI in education” resource provides provincial K-12 guidance within a human-centered approach. It is a useful Canadian counterexample to treating Canada as a single jurisdiction: education authority sits substantially at provincial level, so British Columbia and Quebec should be independent atlas records. citeturn15search0turn15search7

BC: `https://www2.gov.bc.ca/gov/content/education-training/k-12/administration/program-management/ai-in-education`

**Quebec. Original source language: French.** Quebec provides *L’utilisation pédagogique, éthique et légale de l'intelligence artificielle générative*, with provincial resources spanning preschool, primary, secondary and other education sectors. Local organizations remain important for tool-selection and operating conditions. That combination of provincial guidance and local implementation belongs explicitly in the authority model. citeturn15search7

Source: `https://www.quebec.ca/education/numerique/intelligence-artificielle/reseau-education/documents-outils-ia`

**Wales.** Hwb guidance instructs schools to address data protection and consider DPIA responsibilities when integrating GenAI. Estyn's October 2025 inspection/research work found adoption was still largely staff-driven and uneven, with perceived benefits but concerns around training, bias, privacy and uncertainty over obligations. The Welsh Government accepted the need for a more coordinated strategic response. citeturn15search2turn15search8turn15search5

Sources:  
`https://hwb.gov.wales/school-improvement-and-leadership/education-digital-standards/generative-artificial-intelligence-in-education`  
`https://estyn.gov.wales/improvement-resources/new-era-how-artificial-intelligence-ai-is-supporting-teaching-and-learning/`

**Jamaica.** Government reporting says the Education Ministry has been developing draft guidelines for AI in schools while training educators, leaders and students. Jamaica's February 2025 national AI task-force recommendations call for AI/coding curriculum, teacher training for responsible AI and assessment, and stronger privacy/ethics governance. These recommendations are not themselves a final school-use code. The correct status is **policy under development plus pilots and general data-protection rules**.

Sources:  
`https://jis.gov.jm/education-ministry-unveils-initiative-to-integrate-ai-and-adaptive-technology-into-school-curriculums/`  
`https://opm.gov.jm/wp-content/uploads/2025/02/National-Artificial-Intelligence-Task-Force-Policy-Recommendations-Final-1.pdf`

**Trinidad and Tobago.** The education system is actively deploying AI-enabled learning and enterprise tools, but its strategic plan treats development of a **National AI in Education Policy** as a distinct future project. That is important evidence against inferring “policy exists” merely because a government has rolled out AI tools. The atlas should record both implementation and the still-developing governance instrument.

Sources:  
`https://www.moe.gov.tt/educationpolicy2023-2027/`  
`https://storage.moe.gov.tt/wpdevelopment/2026/07/National-Strategic-Plan-Ministry-of-Education-2025-2023.pdf`

**Pacific systems.** Fiji's 2024-33 education-policy framework explicitly includes AI, VR and AR solutions, digital learning, digital literacy and cybersecurity, but no detailed public GenAI classroom permission code was located. Tonga's 2025-35 education framework similarly embeds technology without resolving GenAI homework/examination questions. Samoa's located ICT-in-Education Policy covered 2018-23, so it must not be presented as a current 2026 AI policy absent renewal evidence. Papua New Guinea's 2026 sovereign AI strategy creates broad whole-of-government AI and data-governance direction, including education as a sector, but is not equivalent to K-12 classroom guidance.

Sources:  
`https://www.education.gov.fj/wp-content/uploads/2024/02/2023-Denarau-Declration.pdf`  
`https://www.education.gov.fj/?p=14554`  
`https://www.education.gov.to/images/TESFP%20-%20Full%20version%20Final%20-%20Printers%20marks%201.pdf`  
`https://www.mesc.gov.ws/wp-content/uploads/2019/09/MESC-ICT-in-Education_Policy_2018-2023-30.10.2018_FINAL.pdf`  
`https://www.ict.gov.pg/wp-content/uploads/2026/03/National-Sovereign-AI-Strategy-v15.pdf`

**Cross-jurisdiction comparison**

| System | Student-age condition | Teacher control | Assessment | Privacy/procurement | Teacher use | Accessibility | Bindingness/local discretion |
|---|---|---|---|---|---|---|---|
| France | Primary no direct GenAI manipulation; supervised from 4e; more autonomy in lycée | Strong | Unauthorized substitution can be fraud; detectors discouraged | No confidential/personal data; no required personal account | Permitted within framework | Must consider inclusion | National framework, operational guidance citeturn6view0turn6view1 |
| Singapore | P1-3 no assigned direct use; P4-6 supervised; secondary conditional | Strong | National exams prohibit AI; coursework may differ | MOE/commercial-tool safeguards | Supported and trained | Education-specific tools | National operational approach citeturn9view1 |
| Norway | 1-7 generally no access; 8-10 cautious; upper secondary use literacy | Teacher competence before pupil access | Assessment safeguards developing alongside approach | National safeguards expected | Supported | Explicit need-based exception | National recommendation, not best described as statutory ban |
| NYC | Moratorium 2K-8; controlled HS access | Strong | AI cannot grade or decide promotion/graduation etc. | ERMA approval and NY student-data rules | Planning/admin uses subject to approval | IEP/504 and ELL exceptions | District policy citeturn14view1turn14view2 |
| England | No national pupil-age ban beyond product terms/safeguards | Setting decides | JCQ/awarding rules govern formal assessment | Data protection, safeguarding, filtering and product standards | Permitted with responsibility | Local risk assessment | Local choice inside national legal duties citeturn12view0 |
| New Zealand | Tool ages and school policy | School/teacher control | Own-work requirement; external NCEA prohibition | Avoid personal data | Human judgment retained | Locally determined | National guidance + qualification rules citeturn9view0 |
| Denmark | No single national classroom age table located | Local school/teacher role | GenAI prohibited in Folkeskole examinations | Municipality/school GDPR responsibility | Recommended within pedagogical purpose | Local | Local classroom discretion + national exam rule |
| Italy | Not reduced to consumer-tool age rule | Institutional governance | Depends on use/assessment regime | Strong DPIA, GDPR and vendor governance | Institutional uses contemplated | Rights/inclusion emphasized | National guidelines + EU/national law |
| Taiwan | Tool ages/accounts and teacher/parent guidance | Strong | Separate homework/assessment guidance | Privacy emphasized | Permitted within guidance | Needs further extraction | National guidance |
| Kazakhstan | Under-13/model-rule special safeguards | Approved-tool/teacher governance | Academic-integrity and exam restrictions in model rules | White-list and data restrictions | Extensive support | Inclusion use explicitly contemplated | National framework plus locally adopted model rules |
| Australia | No single national age ban | Schools/settings operationalize | Assessment design should state permitted AI | Privacy/data governance/cybersecurity | Supported | Explicit inclusion principle | National framework plus state/local rules citeturn5view1 |
| Ireland | Vendor ages must be checked | Human educator remains final checkpoint | Separate SEC requirements; acknowledgment | Avoid personal/sensitive data | Supported with validation | Ethical/inclusive principles | National guidance, school-policy implementation citeturn10view0turn10view1 |

This comparison suggests a compact atlas classification should use **four separate fields**, not a red/green badge:

| Dimension | Suggested values |
|---|---|
| **Authority / bindingness** | Binding law/regulation; binding assessment rule; administrative directive; official guidance; local policy; pilot condition; commentary |
| **Student-use condition** | Prohibited; generally restricted; supervised only; teacher-authorized; locally determined; permitted with conditions; not addressed |
| **Assessment condition** | Prohibited; permitted only if specified; disclosure required; own-work/authenticity rule; locally determined; not addressed |
| **Implementation status** | Announced; consultation/draft; pilot; partially implemented; generally implemented; superseded/withdrawn; unclear |

A fifth, independent **evidence-confidence label** should describe the evidence rather than the conclusion: **High** means current primary source inspected with authority/scope established; **Moderate** means primary source exists but a consequential field or version remains unresolved; **Low** means only partial or secondary evidence is available; **Unknown** means the claim should remain unpublished.

## Policy evolution, implementation evidence, and news

The timeline shows three broad phases. Late 2022 and 2023 were dominated by emergency responses, academic-integrity concern and early guidance. During 2024-25, governments moved toward frameworks, AI literacy and procurement/privacy controls. In 2026, several systems became markedly more specific about age, assessment, approved tools and deployment.

| Date | Jurisdiction | Event and date type | Change | Present interpretation | Source / confidence |
|---|---|---|---|---|---|
| Mar. 29, 2023 | England | Publication | DfE issued national GenAI education guidance | Continues, updated Aug. 12, 2025 | citeturn12view0 **High** |
| Jul. 2023 | Japan | Publication | Provisional MEXT GenAI guidance | Superseded/revised by v2.0 | citeturn4view2 **High** |
| Nov. 17, 2023 | Australia | Publication | National GenAI framework published | Framework retained and subsequently reviewed | citeturn4view1 **High** |
| Jan. 2024 | Washington | Publication | OSPI human-centered AI guidance | Later 2024 editions/updates followed | citeturn13search0 **High-moderate** |
| Dec. 26, 2024 | Japan | Revision | MEXT released v2.0 | Current verified version in this pass | citeturn4view2 **High** |
| Jan. 2025 | Saudi Arabia | Issuance | MOE/SDAIA school GenAI guide | Current dedicated national reference located | citeturn22search1 **Moderate-high** |
| May 12, 2025 | Argentina | Program launch | PaideIA national AI education program | Progressive national implementation | citeturn21view1 **High** |
| Jun. 13, 2025 | France | Framework publication | National age-banded AI framework | Current detailed framework | citeturn4view3 **High** |
| Jun. 2025 | Australia | Review endorsement | Ministers endorsed 2024 Framework Review | Update history, not evidence of an entirely new permission regime | citeturn4view1 **High** |
| Aug. 2025 | Italy | Adoption | National AI-in-schools guidelines adopted | Current national governance framework | Official MIM sources above, **high-moderate** |
| Oct. 1, 2025 | Ireland | Guidance publication | National AI-in-schools guidance | Explicitly living/reviewable | citeturn9view3 **High** |
| 2025-26 school year | Kazakhstan | Implementation | AI elements enter digital literacy/informatics and broader national framework develops | System-wide implementation accelerating | Kazakhstan Ministry, **moderate-high** |
| Jan. 19, 2026 | England | Update | Product-safety expectations updated | Adds child-development, safety and manipulation safeguards | citeturn12view1 **High** |
| Feb. 2026 | Taiwan | Revision/notice | v2.1 primary/secondary GenAI guidance identified | Current ministry portal version | citeturn28search0turn28search14 **Moderate-high** |
| Mar. 12, 2026 | Indonesia | Government decision | Seven ministries establish AI/digital education governance framework | Broader than curriculum policy alone | Official Indonesian source above, **moderate-high** |
| May 6, 2026 | Singapore | Parliamentary clarification | MOE states precise P1-3, P4-6, secondary and examination conditions | Strong current national statement | citeturn9view1 **High** |
| Jun. 19, 2026 | Norway | Government announcement | New age-banded national approach | 1-7 generally shielded from direct AI; cautious progression thereafter | Government source above, **high-moderate** |
| Jun. 25, 2026 | Netherlands | Parliamentary status | National foundational-education digitalization plan, including AI, still being developed | AI Act/procurement guidance already relevant; broader plan pending | Official parliamentary source, **moderate-high** |
| Sep. 2, 2026 | NYC | Announcement/current-year policy | 2K-8 student-facing GenAI moratorium and controlled HS strategy | Current 2026-27 policy | citeturn14view1turn25news29 **High** |

**Implementation and outcome evidence**

The evidence base does **not** support ranking jurisdictions by “AI-policy effectiveness.” England's current DfE guidance explicitly says evidence remains limited on effects on learner development, educational outcomes and safety. Ireland likewise describes the research/practice base as early and its guidance as an initial, reviewable step. citeturn12view0turn9view3

| Evidence | Population/setting | Design | Finding useful to the atlas | Limitation |
|---|---|---|---|---|
| Estyn, *New era: how artificial intelligence is supporting teaching and learning*, Wales, Oct. 2025 | Welsh schools, including varied settings | Inspectorate thematic work | Adoption was uneven and often staff-driven; schools reported opportunities while needing clearer training, privacy and governance support. citeturn15search8 | Not a causal study of learning gains |
| Welsh Government response, Oct. 21, 2025 | System level | Government response to inspectorate evidence | Government accepted the need for a more coordinated strategic approach. citeturn15search5 | Policy response, not outcome evaluation |
| England DfE current GenAI guidance | Schools/colleges | Evidence-informed government guidance | Government explicitly says learner/safety evidence is limited. citeturn12view0 | Not an empirical trial |
| Ireland AI-in-Schools guidance | Schools | Evidence-informed guidance | Treats AI-in-education research/practice as emerging and commits to review. citeturn9view3 | Not an outcome study |
| Singapore SG-LEADS announcement/context | Singapore learners | Planned longitudinal research | MOE is building longitudinal evidence rather than claiming established effects. citeturn9view1 | Data collection was described as future work, so there are no outcome results to report yet |
| Pacific Community regional work | Pacific education systems | Regional policy/evidence synthesis | Highlights infrastructure, equity, cultural context, teacher capability and data governance as unusually consequential Pacific implementation factors. | Regional synthesis cannot substitute for country-specific policy or causal evaluation |

**Critical evidence conclusion:** the atlas can confidently compare **what governments require and how implementation is organized** much sooner than it can compare **whether one policy improves learning more than another**. Self-reported teacher time savings, vendor claims, adoption counts and pilot launches should therefore never be displayed as learning-effectiveness evidence.

**Verified news/update dataset**

| Headline / event | Publisher | Publication or event date | Jurisdiction | Category | Why it matters | Policy change? |
|---|---|---:|---|---|---|---|
| NYC adopts 2K-8 GenAI moratorium and controlled high-school approach | NYCPS; AP supporting report | Sep. 2, 2026 | New York City | Restriction + literacy/pilots | One of the clearest recent US reversals from broad experimentation toward age-banded controls | **Yes, operational policy** citeturn14view1turn25news29 |
| Quebec updates provincial digital/AI education resources | Quebec education authorities | Aug. 2026 activity | Quebec | Guidance / literacy | Demonstrates province-level rather than “Canada-wide” authority | **Guidance development**, not national law citeturn15search7 |
| Netherlands reports progress on national digitalization plan for foundational education | Dutch government/parliament | Jun. 25, 2026 | Netherlands | Governance | AI is being integrated into wider public-value and digital-safety governance | **Not yet final national plan** |
| Norway announces age-adapted national recommendations | Norwegian Government | Jun. 19, 2026 | Norway | Age restrictions | Grades 1-7 generally shielded from direct AI; staged access thereafter | **National policy direction/recommendation** |
| European Commission updates ethical AI/data guidance for educators | European Commission | Jun. 9, 2026 | EU | Educator guidance | Updates primary/secondary educator guidance for the AI Act era | **Guidance, not national permission law** |
| Trinidad and Tobago ministries formalize FutureReadyTT education partnership | Government/TTT | May 28, 2026 | Trinidad and Tobago | Procurement / implementation | Demonstrates deployment can precede completion of a national AI-in-Education policy | **Implementation, not final policy code** |
| Singapore clarifies school AI conditions | MOE parliamentary reply | May 6, 2026 | Singapore | Student use / assessment | Precise national age and examination rules | **Authoritative clarification** citeturn9view1 |
| Indonesia adopts seven-ministry AI/digital education framework | Government of Indonesia | Mar. 12, 2026 | Indonesia | Governance | Broadens AI governance across education levels and actors | **Yes** |
| Taiwan advances v2.1 GenAI school guidance | Taiwan MOE | Feb. 2026 | Taiwan | Guidance revision | Shows rapid versioning and assessment-specific guidance | **Revision** citeturn28search0turn28search14 |
| England updates GenAI product-safety standards | Department for Education | Jan. 19, 2026 | England | Procurement / child safety | Extends the governance layer beyond teacher guidance to product design and child-development risk | **Guidance update** citeturn12view1 |
| Ireland publishes national AI-in-Schools guidance | Department of Education and Youth | Oct. 1, 2025 | Ireland | National guidance | Establishes privacy, human-review, age and policy-update expectations | **Yes, guidance** citeturn9view3 |
| Wales inspectorate reports on actual AI adoption | Estyn | Oct. 9, 2025 | Wales | Implementation evidence | Separates policy intent from actual school practice | **No, evidence/reporting** citeturn15search8 |

Press releases, inspection findings and news reports should all remain separate from `policy_documents`. A news row may point to a policy event, but it must never silently become the evidence for the rule when the primary document exists.

## Practical classroom scenario guide

| Scenario | Questions that determine the answer | Evidence-backed comparison | Practical classification |
|---|---|---|---|
| **A 10-year-old uses a public chatbot for homework** | What grade? Is the tool school-approved? Does its account age permit use? Teacher assignment or independent home use? Parent consent? Personal data? | In Singapore, P1-3 are not assigned work requiring direct AI and P4-6 use is supervised with education-specific tools. France does not provide direct GenAI manipulation at primary level. Norway's 2026 approach says grades 1-7 generally should not receive AI access. England has no equivalent blanket age ban but requires product-age compliance and safeguards. citeturn9view1turn6view0turn12view0 | **No universal answer.** Age, tool and supervision are decisive. |
| **A teacher uploads identifiable student essays to a public chatbot for feedback** | Is personal data necessary? Who is controller/processor? Does vendor train on inputs? Has tool been approved? DPIA? | England recommends avoiding personal data and requires legal compliance if use is strictly necessary. Ireland says unclear-input data should effectively be treated as public and personal/sensitive material avoided. Italy puts strong DPIA/vendor duties on institutional AI deployment. New Zealand advises against personal-data submission. citeturn12view0turn10view0turn9view0 | **Usually unacceptable without institutional approval and a lawful, assessed data arrangement.** “The AI can technically accept the essay” is irrelevant. |
| **A secondary student asks AI to outline assessed coursework** | Does the assessment permit AI? Is outlining distinguished from generating submitted content? Must use be disclosed? Can authorship be demonstrated? | Singapore permits AI in some coursework only where consistent with assessment objectives and monitoring, while national exams prohibit it. New Zealand requires the assessed evidence to be the student's own. France treats unauthorized substitution for the student's intellectual work as fraud. | **Task-specific permission is essential.** An outline may be permissible in one assessment and misconduct in another. citeturn9view1turn9view0turn6view1 |
| **A school asks an AI system to generate grades** | Is AI recommending or deciding? Can the teacher meaningfully review? Is the use high-risk under applicable AI law? Does qualification policy permit it? | NYC explicitly prohibits AI from grading or making consequential student decisions. New Zealand stresses human professional judgment. Italy's governance framework requires human-centered oversight and additional assessment for higher-risk systems. citeturn14view2turn9view0 | **Do not equate feedback assistance with automated grading authority.** |
| **A teacher translates lesson materials using AI** | Any student names or confidential text in the prompt? Is translation checked by a competent human? Accessibility need? Approved tool? | England permits staff use within responsibilities but warns about data. Australia emphasizes human accountability and inclusion. Ireland requires human validation of outputs. citeturn12view0turn5view1turn10view0 | Usually among the more defensible staff uses when data are controlled and output is reviewed. |
| **A learner uses AI as an accessibility accommodation** | Is AI named or consistent with an IEP/504/equivalent plan? Does an age restriction conflict? Does the tool process sensitive disability data? | NYC's 2K-8 moratorium expressly preserves qualifying IEP/504 assistive technology and ELL support. Norway's age-based recommendation also contemplates need-based AI exceptions. Australia places accessibility and inclusion inside its framework. citeturn14view1turn5view1 | A general age restriction does **not** automatically eliminate accommodation exceptions. |
| **A school purchases an AI tutor** | Who controls data? Training use? Retention/deletion? Age assurance? Safety filtering? Human escalation? Vendor contract? Automated profiling? | England adds product-safety standards; Italy requires serious controller/vendor/DPIA governance; Australia requires privacy/data controls; the Netherlands' SIVON/Kennisnet framework translates AI Act risk questions into school procurement. citeturn12view1turn5view1 | Procurement is a policy question independent of whether teachers are allowed to “use AI.” |
| **A student is accused based on an AI detector** | Is detector output the sole evidence? What other authorship evidence exists? Is the student offered an explanation/appeal/oral defense? | France expressly discourages AI detectors because of unreliability. Kazakhstan's Academy model rules say detector output should be advisory rather than sole conclusive proof and provide for an oral defense/review process. | **Detector score alone is weak evidence.** Procedural fairness should be a dedicated atlas field. |
| **A teacher generates a worksheet, checks every item, then uses it in class** | Approved tool? Personal/confidential input? Copyright? Accuracy? Age appropriateness? Teacher accountable? | Australia, England and Ireland all support teacher productivity uses while retaining human responsibility for the resulting material. citeturn5view1turn12view0turn10view0 | Commonly permissible subject to tool/data policy and human checking. |
| **A parent asks whether a school-approved tool trains on student data** | What do the contract and privacy notice say? Model training on prompts? Retention? Subprocessors? Cross-border transfer? Can training be disabled? | Singapore distinguishes MOE-built environments and commercial tools and requires safeguards around identifiable information. Australia explicitly treats data handling, retention and onward use as governance issues. England requires transparency about pupil data processing. citeturn9view1turn5view1turn12view0 | “School approved” is not an answer. The vendor/data-processing terms are the evidence. |

These scenarios should become a second navigation layer in EduPolicyAtlas. A parent usually starts with “Can my child use ChatGPT?”, not “What is the administrative bindingness classification of document X?” The scenario page can ask the necessary factual questions and then resolve them through the underlying structured rules.

## Research method, monitoring, community review, and structured data

The atlas's strongest maintainable research model is **field-level evidence**, not document-level citation. A document can be authoritative yet fail to support a particular claim. Every consequential public field should therefore point to a source ID and locator independently.

The recommended lifecycle is:

**Discovered → source authenticated → extracted → translated if needed → rule-level verified → conflict checked → editorially approved → published → monitored → superseded/corrected.**

Automated link checking should never silently advance a source from “current” to “verified.” A 200 response proves only that a URL returned content. It does not prove the document is still controlling, unchanged, or correctly interpreted.

**Date handling should separate at least:** `publication_date`, `revision_date`, `announcement_date`, `effective_date`, `superseded_date`, `source_retrieved_at`, and `rule_verified_at`. Never turn “June 2025” into `2025-06-01` simply to satisfy a database field.

**Translations should record:** source language, original title, English rendering, translation method (`original-language interpretation`, `official translation`, `machine translation`, `human-reviewed translation`), and interpretation notes. “Official English translation” should be reserved for an issuing authority's translation. Japan is a useful test case because MEXT itself labels its English material tentatively. citeturn4view2

**Authority conflicts should resolve downward to the practical layer:** law or regulation can constrain ministry guidance; ministry guidance can coexist with examination-board rules; district policy can narrow national discretion; a school's local policy can further restrict use where higher-level rules permit local choice. Vendor terms are a separate constraint and must never be described as government policy.

**Monitoring watchlist**

| Priority source | Jurisdiction | URL | Language | Monitor | Frequency | RSS/API status from this pass | Human review trigger |
|---|---|---|---|---|---|---|---|
| Australian Dept. of Education GenAI framework | Australia | `https://www.education.gov.au/schooling/resources/australian-framework-generative-artificial-intelligence-ai-schools` | English | Version/review notices | Weekly | Not verified | Any revised PDF, ministerial endorsement or effective-date change |
| MEXT GenAI portal | Japan | `https://www.mext.go.jp/a_menu/other/mext_02412.html` | Japanese | Guideline versions, translations, pilots | Weekly | Not verified | New version number or implementation notice |
| French Ministry AI framework | France | `https://www.education.gouv.fr/cadre-d-usage-de-l-ia-en-education-450647` | French | Age rules, Pix requirements, assessment wording | Weekly | Not verified | Changed age/assessment/account language |
| DfE GenAI guidance | England | `https://www.gov.uk/government/publications/generative-artificial-intelligence-in-education/generative-artificial-intelligence-ai-in-education` | English | Updated guidance | Weekly | GOV.UK change history available; API/RSS not verified here | Material “Updated” entry |
| DfE GenAI product safety | England | `https://www.gov.uk/government/publications/generative-ai-product-safety-standards` | English | Product/child-safety rules | Weekly | Same | New safety expectation |
| NZ Ministry GenAI | New Zealand | `https://www.education.govt.nz/school/digital-technology/generative-ai` | English | Classroom/privacy guidance | Weekly | Not verified | Assessment or age/tool change |
| Singapore MOE | Singapore | `https://www.moe.gov.sg/news` | English | Parliamentary replies, SLS tools, assessment | Weekly | Feed/API not verified | New pupil-age, assessment or approved-tool statement |
| Chile Ciudadanía Digital AI | Chile | `https://ciudadaniadigital.mineduc.cl/ia/` | Spanish | Guidance/resources | Biweekly | Not verified | New national guide |
| Ireland AI-in-Schools guidance | Ireland | `https://www.gov.ie/` plus current guidance URL | English | Living guidance revisions | Weekly | Not verified | Replacement PDF/version |
| OSPI human-centered AI | Washington | `https://ospi.k12.wa.us/student-success/resources-subject-area/human-centered-artificial-intelligence-schools` | English | Edition changes | Weekly | Not verified | New edition/document |
| NYCPS AI guidance | NYC | `https://www.schools.nyc.gov/about-us/policies/guidance-on-artificial-intelligence` | English | Grade restrictions/pilots | Weekly | Not verified | Moratorium, pilot or approved-tool change |
| Taiwan MOE AI download portal | Taiwan | `https://pads.moe.edu.tw/download.php` | Traditional Chinese | Version numbers | Weekly | Not verified | New v2.x/v3.x |
| Norwegian Government / Udir | Norway | `https://www.regjeringen.no/` and Udir AI guidance | Norwegian | Final age recommendation, assessment | Weekly during rollout | Not verified | Final Udir text or revision |
| Denmark UVM | Denmark | `https://uvm.dk/` | Danish | Teaching/exam AI guidance | Weekly | Not verified | Examination-regulation amendment |
| Italian MIM / Unica | Italy | `https://www.mim.gov.it/` | Italian | Guideline/model updates | Weekly | Not verified | Ministerial decree or updated guideline |
| Indonesia Kemendikdasmen/JDIH | Indonesia | Official education/JDIH sources above | Indonesian | Joint decisions, curriculum, safeguards | Weekly | Not verified | New legal instrument |
| Kazakhstan Ministry/Academy | Kazakhstan | `https://www.gov.kz/memleket/entities/edu/` and `https://uba.edu.kz/` | Kazakh/Russian | National standards and model rules | Weekly | Not verified | Final standards, legal-status changes |
| Quebec education AI portal | Quebec | `https://www.quebec.ca/education/numerique/intelligence-artificielle/reseau-education/documents-outils-ia` | French | Guide revisions | Biweekly | Not verified | Revised guide/tool requirements |
| Hwb / Estyn | Wales | URLs above | English/Welsh | Guidance + implementation evidence | Biweekly | Not verified | New national policy/inspection report |
| Saudi Ministry of Education | Saudi Arabia | `https://www.moe.gov.sa/` | Arabic | Guide revisions | Biweekly | Not verified | New national AI/school directive |

Because RSS/API availability was **not separately verified for most sources**, the atlas should not claim these are automated feeds. The safe initial architecture is a scheduled URL/change monitor plus a mandatory human-review queue for semantic changes.

A monitored change should create a **revised policy record** when controlling text or scope changes; a **timeline event** when the change materially shifts practical rules; a **news item** when it is publicly newsworthy; a **correction notice** when prior atlas content was wrong at the time it was published; and a **pending-review flag** whenever the system detects changed source content that has not yet been interpreted.

**Community-submission rubric**

| Review question | Accept when | Reject / hold when |
|---|---|---|
| Is the source authentic? | Issuing authority, official repository or independently verifiable publication | Screenshots without provenance, anonymous claims, fabricated/redirected URLs |
| Is jurisdiction/scope established? | Authority and affected schools/ages can be identified | Submitter extrapolates a school rule to a country |
| Is it current? | Version/current status checked | Clearly superseded or undated with unresolved successor |
| Policy or news? | Correctly classified | Reporting is presented as the governing rule |
| Is interpretation faithful? | Summary tracks actual text and conditions | “AI banned” where source says “not recommended,” for example |
| Translation acceptable? | Original retained; method disclosed | Machine translation represented as official |
| Duplicate? | Adds new document/rule/version | Same source already captured without new information |
| Privacy/sensitive content? | Public institutional information only | Private student, family or employee data |
| Allegations? | Supported by authoritative public evidence | Unsupported misconduct or safety accusation |
| Conflict of interest? | Disclosed where relevant | Vendor marketing submitted as independent evaluation |
| Public summary quality | Neutral, precise, scoped and citation-ready | Advocacy language presented as established fact |
| Correction/withdrawal | Source clearly supports correction | Attempt to erase a valid historical event without evidence |

A lightweight reviewer form should ask: **What authority issued this? What exact jurisdiction and school population does it cover? Is it current? Is it policy, guidance, examination rule, news, research or commentary? What sentence or section supports the proposed rule? Is the source in the original language? How was it translated? What earlier atlas record would this change?**

**Structured-data package**

No downloadable files were written before the research execution window closed. The research itself can nevertheless be normalized directly into the requested files. The recommended relationships are:

```text
jurisdictions.csv
  jurisdiction_id PK
  parent_jurisdiction_id FK nullable
  country_or_area
  jurisdiction_name
  administrative_level
  education_authority
  iso_or_m49_reference
  research_status
  verification_date

policy_documents.csv
  document_id PK
  jurisdiction_id FK
  authority
  original_title
  english_title
  source_language
  document_type
  bindingness
  publication_date
  revision_date
  effective_date
  superseded_date
  current_status
  translation_status

policy_rules.csv
  rule_id PK
  document_id FK
  jurisdiction_id FK
  rule_dimension
  population
  age_min
  age_max
  grades
  school_types
  rule_summary
  conditions
  exceptions
  authority_level
  exact_locator
  source_id FK
  verification_date
  evidence_limitation

sources.csv
  source_id PK
  source_type
  issuing_body
  title
  url
  language
  publication_date
  retrieved_at
  primary_or_secondary
  access_status
  archive_url
  notes

policy_events.csv
  event_id PK
  jurisdiction_id FK
  document_id FK nullable
  event_date
  date_type
  announcement_date
  effective_date
  event_type
  previous_position
  new_position
  source_id
  confidence
  unresolved_questions

news.csv
  news_id PK
  jurisdiction_id FK
  headline
  publisher
  publication_date
  event_date
  category
  summary
  url
  primary_source_id
  related_document_ids
  establishes_policy_change
  dedupe_key

research_evidence.csv
  evidence_id PK
  jurisdiction_id FK nullable
  title
  authors_or_body
  population
  education_level
  sample_size
  design
  intervention
  comparison
  data_collection_dates
  findings
  limitations
  funding_conflicts
  peer_review_status
  policy_relevance
  source_id

coverage_gaps.csv
  gap_id PK
  jurisdiction_id FK
  gap_dimension
  status
  searches_completed
  missing_evidence
  next_authority_to_check
  priority

monitoring_sources.csv
  monitor_id PK
  organization
  jurisdiction_id
  url
  language
  content_type
  frequency
  rss_api_status
  meaningful_change_definition
  automation_suitability
  human_review_required

existing_record_corrections.csv
  correction_id PK
  jurisdiction_id
  existing_statement
  finding
  proposed_replacement
  source_ids
  reason
  remaining_uncertainty
  editorial_status
```

`atlas_research.json` should preserve the same IDs and expose nested views for the website without making the JSON itself the canonical database. Unknown values should be `null`, not `"unknown"` or empty strings. “Not applicable,” “not addressed in this document,” “not located,” and “not verified” should remain separate enumerated states.

The accompanying `README.md` should document enumerations, dates, translation rules, primary-key patterns, source authority hierarchy, evidence confidence and the invariant that **a jurisdiction is not a policy document, and a policy document is not a rule**.

## Publication recommendations and open limitations

**Ready for editorial consideration**

The following findings are sufficiently supported for atlas editorial review now:

| Priority | Addition/correction | Why it belongs first |
|---:|---|---|
| 1 | Correct NYC from vague “restricted through grade 8” to the 2026-27 2K-8 student-facing GenAI moratorium, exceptions, HS literacy/pilots and prohibition on AI grading/consequential decisions | Very current, highly actionable, strong primary evidence. citeturn14view1turn14view2turn14view3 |
| 2 | Expand France into precise age, account, data, assessment and AI-detector rules | Turns a broad label into answers parents and teachers can actually use. citeturn6view0turn6view1turn6view2 |
| 3 | Expand Singapore with P1-3, P4-6, secondary, national-exam, disclosure and privacy distinctions | One of the clearest structured national models. citeturn9view1 |
| 4 | Expand New Zealand assessment/authenticity record | Directly answers “Can a student use it on assessed work?” citeturn9view0 |
| 5 | Reframe England as “local discretion inside national legal/safety guardrails” and add Jan. 2026 product standards | Prevents an important misleading inference. citeturn12view0turn12view1 |
| 6 | Add Norway | Strong age-specific 2026 policy direction and accessibility exception |
| 7 | Add Denmark | Excellent example of local classroom discretion coexisting with a national examination prohibition |
| 8 | Add Taiwan | Strong multilingual expansion plus separate general-use and assessment guidance. citeturn28search0turn28search12 |
| 9 | Add Italy | Gives the atlas a detailed institutional governance/privacy/procurement model |
| 10 | Add Kazakhstan, with careful authority labels | Substantially improves Central Asian coverage and exposes how national frameworks and local model rules interact |

Australia, Ireland, British Columbia, Quebec, Argentina and Wales are also publication-ready at the overview level, subject to normal editorial review. citeturn4view1turn9view3turn15search0turn15search7turn21view1turn15search2

**Needs verification or interpretation**

Malaysia's 2026 ministry guide is highly promising but should receive page-level source verification before granular public rules are quoted. Indonesia's seven-ministry framework is strong primary evidence, but the reported red/yellow/green assessment taxonomy should be checked word-for-word against the controlling document before publication. Saudi Arabia's guide should receive detailed Arabic extraction for age, assessment and privacy fields. Chile needs a deeper Spanish rule audit. Washington State needs a final version-history reconciliation. Jamaica's school-specific guidance remains described as draft in the evidence located. Trinidad and Tobago is implementing AI at scale while its strategic plan separately calls for development of a National AI in Education Policy. Samoa needs confirmation of what replaced or extended its 2018-23 ICT policy.

Africa remains the largest geographic evidence gap. Rwanda, Nigeria and South Africa have meaningful AI strategy, curriculum, research or pilot activity, but this pass did not establish sufficiently detailed national K-12 GenAI-use codes. Kenya needs a full Ministry/KNEC investigation. South Asia outside the limited India/Sri Lanka discovery pass is also incomplete. Southeast Asian coverage beyond Indonesia and Malaysia requires deeper original-language work in the Philippines, Thailand, Vietnam, Brunei and Cambodia.

**Do not publish as factual claims**

Do not publish “Country X has no AI policy” merely because one was not found. Do not call Norway's national recommendation a criminal, statutory or absolute “AI ban.” Do not treat Denmark's local classroom discretion as permission to use AI in national examinations. Do not describe Trinidad and Tobago's AI-tool deployment as proof that its planned National AI in Education Policy is already in force. Do not extrapolate a provincial Canadian rule to Canada. Do not infer a ministry ban from a vendor's age restriction. Do not infer student permission from a teacher-training program. Do not state that an AI detector proves misconduct. Do not describe a pilot as system-wide authorization. Do not rank countries by educational effectiveness based on the evidence gathered here.

**Open limitations**

The most material unfinished work is the full original-language, page-level extraction for Japan, Chile, Malaysia, Saudi Arabia and selected Indonesia documents; examination-authority audits outside the strongest existing records; systematic state/province/Land/district sampling in decentralized systems; a deeper Sub-Saharan African pass; South Asia; additional Southeast Asian languages; Caribbean CXC implementation; and a larger empirical K-12 research review.

The original raw research-report Markdown and any current policy/news export files were not available in the materials I could inspect, so this package does not claim to have reconciled hidden atlas records against those exports. The live HTML site and report were inspected, and corrections are framed accordingly. citeturn1view0turn2view0

The strongest immediate product decision is to make **rule-level comparison the core data model**: age/access, teacher authorization, assessment, data/privacy, accessibility, staff use, bindingness and implementation should each be independently filterable and independently cited. That architectural change would do more to reduce parent and educator confusion than adding another thirty country cards with a single “allowed/restricted” label.

**First action:** update the NYC, France, Singapore, New Zealand and England records before expanding the public map, because those five corrections add more practical user value with stronger evidence than a larger batch of shallow new-country summaries.

[2026-09-08]