The European Commission announced on 22 September 2026 a paper that reviews the evidence on generative AI in learning and teaching, under the title Friend or foe [1]. A key finding is that generative AI can improve how students perform on a task, but that this does not necessarily mean they learn more [1].
The paper also reports that four in ten young people aged 16 to 24 in the EU used generative AI tools for formal education in 2025, citing Eurostat data [1]. It calls for more European research to understand the long-term effects [1].
What did the European Commission publish?
The Commission released two reports together, according to its news announcement [1]. The first is the Friend or foe paper on generative AI. The second is a Eurydice report on digital education at school in Europe in 2026 [1].
The AI paper reviews scientific evidence and international studies [1]. It covers students, teachers, parents and the wider education system [1]. Topics include cognitive development, critical thinking, self-regulated learning, equity and socio-emotional effects [1].
What does the AI and learning evidence say about performance?
The Commission summarises it plainly: "AI can improve performance, but this does not necessarily mean better learning" [1].
This is the key distinction for teachers. A student can hand in a better essay or solve more problems with AI help. That does not show the student has understood more, or could do it again without the tool.
According to the summary, students are more likely to benefit when AI supports their own thinking and they engage actively with the content [1]. The paper also finds that educational AI tools generally give better learning outcomes than general-purpose tools [1].
Could AI weaken critical thinking?
The paper raises concerns about over-reliance [1]. Students may use AI to do tasks that are important for building knowledge and cognitive skills [1]. If that happens often, the development of critical thinking could suffer, the summary says [1].
The wording is cautious. It is a concern raised by the review, not a measured long-term effect.
What does it mean for teachers' own work?
The paper says AI could help teachers prepare lesson plans, materials and assignments [1]. It also says using AI well and responsibly puts new demands on teachers [1].
Training is thin. Only 29.7% of EU teachers surveyed in TALIS 2024 had taken part in professional development related to AI for teaching and learning [1].
What role do parents play?
Parents remain an important part of children's learning, the paper finds [1]. Better digital literacy can help parents guide children through digital environments and AI use [1]. The Commission's summary states that "AI cannot replace parental guidance and support" [1].
What does the companion Eurydice report add?
The Eurydice report draws on data from 38 education systems in the Eurydice Network and 24 systems in ICILS 2023 [1]. It finds that nearly two-thirds of education systems now address AI through national strategies, guidance or policy frameworks [1]. Access to advanced AI technologies at school level remains uneven [1].
It also finds gaps in teacher support, especially in personalised learning, assessing online information and the use of AI [1]. Digital competence is taught in lower-secondary education in most systems, with around three-quarters offering a dedicated subject [1].
Who is affected?
- Teachers: most directly, through task design, assessment and the training gap: only 29.7% of EU teachers surveyed in TALIS 2024 had AI-related professional development [1].
- Students: four in ten young people aged 16 to 24 in the EU used generative AI for formal education in 2025 [1].
- School leaders: the evidence favours educational tools over general-purpose ones [1], which may affect purchasing and training plans.
- Parents: the paper keeps them central to children's learning [1].
- Trainers and facilitators of remote teams: the review concerns education [1]. Applying its findings to adult workshops would require separate evidence.
What is not yet clear?
- Long-term effects. The paper itself calls for large-scale, representative and longitudinal research in European education contexts [1]. Today's evidence does not show what AI use does to skills over years.
- Effect sizes. The Commission's summary does not give figures for how much performance rises or how much learning changes.
- Which tools count as educational. The summary contrasts educational and general-purpose tools [1] but does not define the line.
- Country differences. Use differs a lot between countries [1], but the summary gives no country figures.
- Policy follow-up. The findings will feed into a forthcoming Education Package on basic skills, teachers and trainers, and digital infrastructure [1]. The source gives no date and no detail beyond these three topics.
What can teachers do now?
- Separate the product from the learning. Ask what a task is meant to teach. If AI does that part, check separately what the student can explain or do unaided.
- Have students think first. Ask for a first attempt, a plan or a prediction before any AI use. The review links benefit to AI supporting students' own thinking [1].
- Assess the reasoning. Consider short oral checks, drafts or reflections alongside a finished product to see how students reason.
- Prefer tools designed for learning where you can. The evidence generally favours educational AI tools [1].
- Look for training. With only 29.7% of surveyed EU teachers having taken part in AI-related professional development [1], ask your school or authority what is on offer.
- Bring parents in. Share your class rules on AI so guidance at home matches guidance at school.
- Read the source. The full paper is linked from the Commission's announcement [1].