The day AI policy turned into operating rules.

This issue’s through-line is governance becoming operational: who can use AI, when, how it is assessed, and how teachers are trained.

1Top story

Brazil · National guidance

Brazil moves AI in education from adoption to risk governance.

Brazil’s National Education Council approved national guidelines for AI use across schools and universities, a notable step because it treats education AI as a risk-governance problem rather than a procurement trend. Reporting from Porvir, Correio Braziliense, and Tecnoblog says the rules create risk categories and limits on sensitive uses such as automated essay correction, with institutions expected to adapt after ministry homologation.

Brazil is moving the debate from “AI in the classroom” to “which educational decisions are too important to automate.”

2Research & evidence

The evidence keeps favoring guided use.

AI tutors can help with math — if they hold back.

University of Toronto Mississauga research found AI tutors can support math learning when they withhold answers and require reasoning.

Phys.org

Engagement is not learning.

The American Psychological Association cautioned that engagement data is not the same as learning evidence, especially with generative AI.

Phys.org

Assessment redesign is going global.

Professors worldwide are moving toward process, oral defense, and AI-resistant tasks rather than relying on detectors.

Nature

The offloading concern is still live.

A widely shared report on faster homework but poorer exams kept attention on whether AI supports thinking or replaces it.

Al Jazeera on MSN

3Around the globe

What else moved this week.

🇧🇷 BrazilNational AI-in-education guidelines now cover schools and higher education, with risk categories and limits on automated assessment. Porvir

🇲🇽 MexicoSEP and UNESCO agreed to cooperate on ethical AI, devices, and social-media policy for education. La Jornada

🇦🇪 UAESchools are rolling out an AI curriculum across public and private systems, paired with large-scale teacher training. Khaleej Times

🇯🇵 JapanThe curriculum draft expands information education and embeds AI beyond computing classes. The Mainichi

🇮🇳 IndiaGoogle’s multilingual teacher training and ABP Education’s YoLearn.ai investment both target mass AI-learning access. Education Times

🇵🇰 PakistanPakistan-China cooperation on AI was highlighted as a growth area for education, industry, and public services. Daily Times on MSN

🇦🇺 AustraliaSchool leaders still face an “AI strategy gap,” with adoption moving faster than clear institutional planning. The Educator

4Tools & platforms

Platforms are moving from prompts to tutoring workflows.

ABP Education backs voice-first tutoring in India.

The investment in YoLearn.ai targets textbook-aligned personal AI tutors for Indian students.

Republic World

Google and Khan Academy expand Khanmigo.

Google.org fellows and Gemini added real-time interactive diagrams and a teacher-controlled practice builder.

GCN

Duke and Simplilearn move into agentic AI credentials.

Their new professional certificate covers applied generative AI and agentic AI for adult learners.

TMCnet

5Worth watching

Three patterns to track.

Risk categories are becoming the policy language.

Brazil’s framework, EU-style risk thinking, and district age gates all move beyond broad “AI allowed/banned” rules.

Teacher training is the new adoption layer.

India and the UAE show that curriculum rollout now comes with educator enablement.

Assessment reform is spreading faster than detection.

Higher education is moving toward proof-of-process instead of policing finished text.

AI in Education Daily · Berit Herstad · September 9, 2026 · Compiled from public sources