By Dr Tim Kitchen
With a population of around 1.38 million, slightly smaller than Adelaide’s, Estonia sits on the Baltic Sea in Northern Europe, across the water from Finland and sharing borders with Russia and Latvia. It has 464 schools serving about 163,000 students, with an education system well known for teacher autonomy, equity and digital learning.
In 2025, the Estonian government launched AI Leap (TI-Hüpe) to support critical, responsible and effective AI use in schools. The starting point was a reality familiar to many Australian teachers: students were already using AI, with or without permission (AI Leap Foundation, 2026a, p. 3).
Teachers received access to premium ChatGPT and Google Gemini, along with training and support, before student accounts were introduced in January 2026 (AI Leap Foundation, 2026).
Schools were also given a learning application called ITI, developed with researchers and technology partners. It takes a Socratic approach, using questions, explanations and feedback to guide students towards their own conclusions. During 2026, the program involved 154 of Estonia’s 156 upper-secondary schools, reaching approximately 20,000 students in Years 10 and 11 and 5,000 teachers (e-Estonia, n.d.). By the end of 2027, it is expected to extend to all upper-secondary year levels, vocational students and teachers, and teachers of Years 5 to 9 (AI Leap Foundation, 2026a).
The first annual review, published in August 2026, offers a rare system-wide account of a country taking a deliberate approach to bringing AI into schools and gives Australian educators and leaders something concrete to learn from.
What AI Leap actually did
AI Leap starts with the view that meaningful learning takes active mental effort. The technology needs to support the teaching, with students still doing the thinking. A catchphrase used with teachers summed this up: “take the stairs, not the elevator” (AI Leap Foundation, 2026a, p. 13).
In its first year, the program focused on three areas:
- Teachers. Every upper-secondary teacher was offered a paid ChatGPT or Google Gemini account. Each school nominated two teacher leaders to attend monthly regional seminars and bring that learning back to their colleagues through school-based learning circles, usually a 90-minute meeting each month. Subject associations also helped write materials for maths, history, civics, music and languages (AI Leap Foundation, 2026a, pp. 11–15).
- Students. The program included an introductory AI literacy course and 23 online lessons. A “Be AI Smart!” video campaign brought in perspectives from experts in law, journalism, art and science. Youth organisations were involved in 41 events, ranging from debate tournaments and hackathons to 18 World Café discussions involving around 950 students (AI Leap Foundation, 2026a, pp. 9–10).
- Technology. Working with OpenAI and Estonian researchers, the program developed ITI, a ChatGPT-based study app in Estonian that guides students with questions rather than simply giving them answers. The Institute of the Estonian Language also developed public benchmarks to test AI models on Estonian language, curriculum content and child safety (AI Leap Foundation, 2026a, pp. 17–18).
Funding came from the Ministry of Education and Research, on the condition that the private sector matched half the first-year budget. By May 2026, more than 30 businesses had contributed close to €1 million (AI Leap Foundation, 2026a, p. 38).
Six findings worth reflecting on
1 – AI use among surveyed teachers was already widespread. By early 2026, 95.9% of teachers surveyed had used AI in their work. They estimated it saved them around 1.8 hours a week, most often through creating differentiated materials, planning lessons, preparing feedback and handling admin tasks such as parent emails (AI Leap Foundation, 2026a, pp. 27–28).
2 – Teachers were moving from talking about change to changing their practice. Comparing the late 2025 and early 2026 surveys, there was slightly less emphasis on broad discussions about changing skills, alongside more concrete changes to teaching practice. (AI Leap Foundation, 2026a, pp. 24–25). That is an interesting shift in thinking and practice.
3 – Concerns about students’ thinking remained strong. In the early 2026 survey, 80.2% of teachers agreed that students may be handing their thinking over to AI, while 73.4% said students could produce high-quality work they couldn’t explain. Only 11.7% believed students had developed a habit of checking AI responses against other sources. Access to the tools clearly doesn’t remove the need to teach students how to question what they receive (AI Leap Foundation, 2026a, p. 25).
4 – Learning circles were a particular success. The proportion of schools running AI-focused learning circles grew from 70.5% to 83.4%. Teachers involved reported high motivation and confidence in their abilities, with relatively low stress. The report describes collaborative professional learning as “one of the strongest outcomes of the program”. The questions teacher leaders kept raising will sound familiar to Australian educators: how do we involve sceptical colleagues, where do we find the time, and who sets the rules? (AI Leap Foundation, 2026a, p. 14).
5 – Getting teenagers to use an AI tutor that makes them think isn’t easy. Around 12,400 students activated ITI accounts, but monthly active use peaked at about 5,700, below the program’s 60% target. Average satisfaction was 3.07 out of 5. A recurring complaint, appearing in around 25 comments, was that it asked too many questions. When students became frustrated, many switched to regular ChatGPT or Claude. They reported using other AI tools more often than ITI, which raises a practical challenge for educators trying to encourage deeper thinking through AI (AI Leap Foundation, 2026a, p. 31).
6 – Students who used ITI generally used it to support learning. An analysis of around 218,000 conversations between January and May 2026 found that most active students showed learning-oriented or mixed patterns of use. Only 6.9% mainly sought ready-made answers, and this group used the app much less. The researchers manually set the thresholds for these categories, so the precise percentages need some caution. The useful message for teachers is to look at a student’s overall pattern of AI use. An occasional request for an answer tells us very little on its own (AI Leap Foundation, 2026a, pp. 29–30).
Student’s voice
Between March and May 2026, around 950 students in Years 10 to 12 took part in World Café discussions across 18 Estonian schools. Their recommendations give us plenty to consider here in Australia. Students called for more meaningful learning, clearer expectations around AI and a greater say in their education (Estonian National Youth Council et al., 2026).
Learning
- Set tasks that require analysis, creativity and students’ own perspectives, giving them a reason to think for themselves.
- Put less emphasis on grades and use a wider range of ways for students to show their understanding, including oral responses, in-class assessments and creative work.
- Have teachers and students develop clear agreements together about when and how AI can be used, and how much assistance is appropriate.
- Give students practical guidance on using AI to build their knowledge and develop independent thinking.
- Explain how learning happens and why knowledge and practice still matter when AI is available.
- Review homework so there are fewer, more meaningful tasks. Students shouldn’t feel they need AI just to get everything finished.
Relationships
- Make room for experimentation in lessons, with opportunities to explore different answers. Treat mistakes as a normal part of learning.
- Provide accessible mental health support and individual feedback. A stronger sense of belonging matters, including when addressing AI misuse.
- Give students a voice in school decisions and support the initiatives they lead.
Creativity
- Be clear about how students should describe and acknowledge their AI use, and how that use will be assessed.
- Have teachers and students openly and consistently label AI-generated content.
- Build foundational skills in writing, creativity, problem-solving and critical thinking during primary school, before students rely heavily on AI.
- Show students how to use AI in creative work while keeping their own ideas and creative process central.
Social media
- Teach age-appropriate media literacy, how algorithms work and the basics of AI through computing or other compulsory subjects.
- Embed critical thinking, discussion, information analysis and source checking across schoolwork.
- Introduce stronger, binding rules for AI and social media platforms, addressing addictive design, transparency and data protection.
Looking ahead
- Update computing classes to cover how AI works, its limitations, data protection and critical evaluation. Include AI learning across subjects in the senior secondary years.
- Give students more say in their learning pathways.
- Connect school learning more closely with everyday life, future skills and career choices.
- Offer opportunities to complete tasks with or without AI. Teach students about the risks and limitations so they can make informed choices.
(Estonian National Youth Council et al., 2026).
How AI Leap connects with Australia
In some respects, Australia is running its own version of Estonia’s experiment, on a state-by-state and sector-by-sector basis. Three trends are particularly similar.
1 – State-built tutors that guide students through questions. Like ITI, several Australian education systems have developed AI tools designed to guide students through their thinking. NSWEduChat opened to NSW public school students from Year 5 in Term 4, 2025 (NSW Department of Education, n.d.; Responsible AI Australia, 2026). InnovationAus describes it as a tool that “does not provide direct answers” (InnovationAus, 2025). South Australia’s EdChat reached public high schools from late 2025 (Furze, 2025), while Education Queensland’s Corella is available to students in Years 7 to 10 with parental consent. The Queensland department describes its approach to the ABC as Socratic, using questions to guide learning (ABC News, 2026).
Estonia’s experience raises a practical challenge here. A tool designed to support learning still needs students to want to use it. If the questioning feels slow or patronising, they’ll simply open another chatbot.
2 – The promise of saving teachers time. Western Australia has committed a further $4.6 million to its ClassmAIte teacher planning pilot, aiming to involve more than 100 schools by the end of 2026 (Education Matters, 2026). Estonian teachers’ estimate that AI saved them around 1.8 hours a week gives us a useful point of comparison. It’s worth remembering, though, that this was self-reported, rather than independently measured (AI Leap Foundation, 2026a, p. 24).
3 – Growing pressure around assessment integrity. A Learning First report commissioned by the NSW Education Standards Authority (NESA), published in May 2026, surveyed almost 3,400 NSW teachers across all sectors. Among teachers who observed student AI use, 73% of senior secondary teachers said students were using it to complete assessment tasks. Around half of secondary teachers said they did not know how to prevent AI-assisted cheating (Williams, 2026). Estonian teachers described a familiar concern: students submitting polished work they couldn’t explain (AI Leap Foundation, 2026a, p. 25).
From Term 4, 2026, NESA’s rules limit unsupervised assessment to 15% of assessment work in Stage 6 Preliminary and HSC courses. Schools must design assessments that demonstrate students’ own knowledge, skills and understanding, ensuring marks do not reflect AI-generated work (NSW Education Standards Authority [NESA], 2026).
The Australian context
These similarities are useful, but the differences between the two countries matter. Estonia has 156 upper-secondary schools, one education ministry and a coordinated national program. Australia has eight jurisdictions and three main school sectors: government, Catholic and independent. Our state-developed AI tools mostly serve government schools.
Many Catholic and independent schools are developing their own approaches. Others, including some in the Steiner network, are deliberately taking a slower approach to AI adoption (ABC News, 2026).
For Australia, one encouraging lesson is the value of giving schools time and support to work things through together. Estonia’s experience suggests that schools’ readiness and opportunities to experiment with colleagues mattered to how well the program worked (AI Leap Foundation, 2026a, p. 39; 2026c).
Six things for Australian teachers, school leaders and policymakers to consider
1 – Involve students in writing the AI rules. This was a clear message from Estonia’s student discussions. Develop a class agreement together about when AI supports learning, when its use must be acknowledged and which tasks need to be completed independently. Include how students will show their own thinking.
2 – Start a small learning circle. Two or three colleagues, an hour a month and one shared question is a manageable starting point. Collaborative professional learning was one of Estonia’s strongest outcomes. If there’s no time set aside for it at your school, that’s a conversation to have with leadership.
3 – Redesign one task to make the reasoning visible. Estonian teachers had students debate an AI, check its claims against primary sources and explain their process in class. One learning circle resource captures this approach in its title: “From Performance to Process: Watch How I Think.”
This connects closely with Principle 3 of the University of Sydney’s Castlereagh Statement green paper, which recommends “process over product” in teaching and assessment. It calls for making learning visible and using AI to support students’ thinking (Castlereagh Summit, 2026).
4 – Teach students how to check AI responses. Only around one in nine Estonian teachers believed students had developed a habit of checking AI output. We shouldn’t assume our students will do this without guidance either. Build source checking into the task, including checking whether a source is credible and actually supports the claim.
5 – Look at patterns of use. A student who occasionally asks AI for an answer isn’t necessarily handing over their learning (AI Leap Foundation, 2026a, p. 29). Look at how they use it across a unit. Talk with them about the choices they’ve made and ask them to explain what they’ve learnt.
6 – Explain why the tutor asks questions. If you use NSW EduChat, South Australia’s EdChat or Queensland’s Corella, talk with students about the purpose of guided questioning. Students in Estonia became frustrated with questions they didn’t see the point of (AI Leap Foundation, 2026a, p. 33). Model how to work through the exchange, and recognise when a direct explanation or worked example would be more useful.
Start with the people
Estonia’s first year of AI Leap shows how much work comes after providing access to the technology. Teachers need time to reconsider tasks and assessment, and students need to understand why doing their own thinking still matters. As the review puts it, “Access to AI alone does not help anyone learn more effectively” (AI Leap Foundation, 2026c).
For me, that’s an encouraging message. We already have considerable expertise in our staffrooms. Giving teachers the time to learn together, try things and talk honestly with students is a practical place to start.
If you would like support (online or in-person) with the safe & ethical integration of AI tools at your school, get in touch with me via – https://timkitchen.net/book-tim/
My new book The Human Advantage – Teaching, Learning and Leading in a Gen AI World is now available to order via – https://timkitchen.net/buy-tims-books/
Have a listen and/or look at my podcast/vodcast – The Kitchen Think on Spotify, Apple Podcasts or YouTube – https://timkitchen.net/podcast/
References
ABC News. (2026, August 17). Why alternative schools are not rushing to bring AI into classrooms. https://www.abc.net.au/news/2026-08-17/qld-alternative-schools-not-rushing-to-bring-ai-into-classrooms/107016124
AI Leap Foundation. (2026a). AI Leap: First annual review 2025/26. https://tihupe.ee/wp-content/uploads/2026/08/AI-Leap-First-Annual-Review-202526.pdf
AI Leap Foundation. (2026b). AI Leap’s first year at a glance 2025/26. https://tihupe.ee/wp-content/uploads/2026/08/AI-Leap-First-Year-short.pdf
AI Leap Foundation. (2026c, August 7). Read the AI Leap first-year report. https://tihupe.ee/en/read-the-ai-leap-first-year-report/
Castlereagh Summit. (2026). The Castlereagh Statement: A cross-sector call to action on Australian education and training in the age of AI. https://castlereagh.ai/wp-content/uploads/2026/04/The-Castlereagh-Statement-v1.00-1.pdf
e-Estonia. (n.d.). AI Leap. https://e-estonia.com/service/e-education-and-research/ai-leap/
Education Matters. (2026). $4.6 million boost extends AI pilot to reduce teacher workload. https://educationmattersmag.com.au/4-6-million-boost-extends-ai-pilot-to-reduce-teacher-workload/
Estonian National Youth Council, Estonian School Student Councils’ Union, & AI Leap Foundation. (2026). What do students think about AI in schools? Summary of the AI Leap World Café discussions. https://tihupe.ee/wp-content/uploads/2026/07/Summary-of-the-AI-Leap-World-Cafe-Discussions_web_.pdf
Furze, L. (2025, December 3). What the National AI Plan means for schools. https://leonfurze.com/2025/12/03/what-the-national-ai-plan-means-for-schools/
InnovationAus. (2025, September 23). NSW rolls out EduChat AI tool to public school students. https://www.innovationaus.com/nsw-rolls-out-educhat-ai-tool-to-all-public-school-students/
NSW Department of Education. (n.d.). NSWEduChat. https://education.nsw.gov.au/teaching-and-learning/education-for-a-changing-world/nsweduchat
NSW Education Standards Authority. (2026, August 31). New limit on take-home assessment tasks from Term 4 2026. NSW Government. https://www.nsw.gov.au/education-and-training/nesa/news/all/new-take-home-assessment-rules
Responsible AI Australia. (2026, October 6). The Australian Framework for Generative AI in Schools, explained: What it requires in 2026 and what each state adds. https://www.responsibleaiaustralia.com.au/news/australian-framework-generative-ai-schools-explained-2026






























































































































































































































































































































































































































































































































































