# The AI Act and the future of STEM education in Europe: rethinking pedagogy, assessment, and teacher agency

- Type: Literature
- Source: Frontiers in Education
- Date: 2026-07-02
- Original: https://doi.org/10.3389/feduc.2026.1845045
- Canonical: https://overview.legal/posts/53829
- Topics: AI Governance Framework, Data Governance for AI, Artificial Intelligence, Right to Explanation, Human Oversight, Monitoring, Transparency, Accountability, Profiling, Automated Decision-Making

## Summary

Generative artificial intelligence (AI) is swiftly transforming STEM education, presenting pedagogical, ethical, and regulatory challenges. The Artificial Intelligence Act, a comprehensive legislative framework emphasizing transparency, accountability, and human oversight, is implemented throughout Europe. This paper analyzes the necessity of re-evaluating STEM education in light of the convergence between generative AI and the AI Act. The paper posits that, grounded in pedagogy, evaluation, and

## Full text

EDITED BY Sevil Akaygun, Bog ̆ aziçi University, Türkiye REVIEWED BY Luiz Antonio Gomes Senna, Rio de Janeiro State University, Brazil Andreea Nicoleta Dragomir, Lucian Blaga University of Sibiu, Romania Senem Çolak Yazici, Duzce University, Türkiye * CORRESPONDENCE Konstantinos T. Kotsis kkotsis@uoi.gr RECEIVED 01 April 2026 REVISED 19 May 2026 ACCEPTED 15 June 2026 PUBLISHED 02 July 2026 CITATION Kotsis KT and Stylos G (2026) The AI Act and the future of STEM education in Europe: rethinking pedagogy, assessment, and teacher agency. Front. Educ. 11:1845045. doi: 10.3389/feduc.2026.1845045 COPYRIGHT © 2026 Kotsis and Stylos. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY) . The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. The AI Act and the future of STEM education in Europe: rethinking pedagogy, assessment, and teacher agency Konstantinos T. Kotsis* and Georgios Stylos Lab of Physics Education and Teaching, Department of Primary Education, University of Ioannina, Ioannina, Greece Generative artificial intelligence (AI) is swiftly transforming STEM education, presenting pedagogical, ethical, and regulatory challenges. The Artificial Intelligence Act, a comprehensive legislative framework emphasizing transparency, accountability, and human oversight, is implemented throughout Europe. This paper analyzes the necessity of re-evaluating STEM education in light of the convergence between generative AI and the AI Act. The paper posits that, grounded in pedagogy, evaluation, and teacher agency, AI functions as an epistemic actor that actively participates in the generation and validation of knowledge. In AI-dominant environments, conventional instructional and evaluation frameworks are inadequate and necessitate new process-oriented, authentic, and ethical methodologies. It also underscores the increasing importance of teachers as mediators of technological and regulatory forces. The interplay among policy, practice, and professional agency is illustrated using a conceptual framework. The paper concludes with implications for educators, teacher training, and policy, contributing to the discourse on European STEM education. KEYWORDS assessment, European AI Act, generative artificial intelligence, STEM education, teacher agency 1 Introduction Generative artificial intelligence (AI) is swiftly transforming STEM education, presenting novel opportunities and challenges. Recent studies indicate that AI technologies are now embedded in core educational processes, influencing how students access information, develop knowledge, and address disciplinary challenges ( Chiu, 2023 ; Fan et al., 2025 ). In STEM disciplines where conceptual understanding, modelling, and problem-solving are essential, the capacity of generative AI systems to elucidate, execute intricate tasks, and replicate scientific reasoning presents significant hurdles regarding learning and proficiency. Although engagement and customized learning may be advantageous, superficial understanding, cognitive offloading, and the erosion of students’ independent problem-solving abilities are considerable issues ( Vieriu, 2025 ). The incorporation of AI into STEM education represents a fundamental shift that interrogates epistemological and pedagogical assumptions, rather than merely a technological progression. The EU’s AI Act is a significant regulatory measure that reshapes the relationship between technology, society, and education within this evolving context. The AI Act TYPE Perspective PUBLISHED 02 July 2026 DOI 10.3389/feduc.2026.1845045 Frontiers in Education 01 frontiersin.org positions Europe as a leader in global AI regulation by establishing a risk-based framework that guarantees transparency, accountability, and fundamental rights ( Veale and Borgesius, 2021 ). The legislation, albeit not intended for educational purposes, has significant implications for teaching and learning. AI systems in education are linked to data governance, algorithmic transparency, and the classification of high-risk applications, particularly in automated decision-making and assessment. Consequently, educators and institutions must navigate a challenging environment where educational innovation must adhere to legal and ethical standards. Although research on AI in education is expanding, it largely concentrates on tool capabilities or immediate teaching outcomes. The systemic implications of integrating AI inside a regulated framework such as the AI Act have garnered minimal scrutiny. The gap significantly impacts STEM education, as disciplinary knowledge is closely linked to epistemic processes such as modelling, argumentation, and evidence-based reasoning. Recent study indicates that generative AI alters student engagement with information and influences the processes of knowledge formation, evaluation, and assessment ( Chiu, 2023 ). Consequently, integrating AI within a legal framework necessitates a comprehensive revaluation of educational methodologies, extending beyond tool implementation to encompass pedagogy and assessment. STEM disciplines present distinctive challenges for AI integration because learning is closely connected to modelling, experimentation, quantitative reasoning, design thinking, and evidence-based problem solving. Unlike many other educational domains, STEM learning depends heavily on epistemic practices through which students construct, test, validate, and apply knowledge claims. Recent research indicates that generative AI systems increasingly participate directly in these processes by generating explanations, solving mathematical problems, proposing experimental procedures, supporting coding tasks, and simulating scientific reasoning ( Chen and Cheung, 2025 ; Huwer et al., 2025 ). Consequently, the implications of AI integration in STEM education extend beyond information access and directly influence how disciplinary understanding and scientific reasoning are developed, evaluated, and validated. These challenges are particularly significant because AI systems often struggle with multistep quantitative reasoning, complex interdisciplinary problem solving, and authentic experimental interpretation, despite their strong performance in text-based interactions ( Chen and Cheung, 2025 ). In engineering and technology education, generative AI tools may support iterative design processes while simultaneously risking over-reliance and reduced cognitive flexibility if students engage passively with AI-generated outputs ( Zhang et al., 2025 ). Similarly, in science education, concerns have emerged regarding “epistemic drift,” where learners may increasingly rely on AI-generated explanations without sufficiently engaging in causal scientific reasoning or empirical validation ( Li et al., 2026 ). These concerns suggest that AI integration in STEM education requires pedagogical approaches that preserve disciplinary inquiry, conceptual understanding, and critical evaluation of AI-generated knowledge. This paper contributes to STEM education research by examining generative artificial intelligence and the European AI Act through the lens of disciplinary epistemic practices rather than general digital learning processes. The paper argues that the implications of generative AI in STEM education are distinctive because STEM disciplines rely heavily on modelling, experimentation, analytical reasoning, design thinking, and evidence-based problem solving. As AI systems increasingly participate in these practices by generating explanations, solving quantitative problems, supporting coding and design tasks, and simulating scientific reasoning, they influence not only access to information but also the production, validation, and assessment of disciplinary knowledge. Within this context, the paper conceptualizes AI as an epistemic actor that reshapes pedagogy, assessment, and teacher agency under emerging regulatory conditions established by the European AI Act. By connecting AI governance with STEM- specific educational practices, the manuscript proposes a conceptual framework for understanding how regulation, pedagogy, assessment, and professional judgment interact in AI- enhanced STEM learning environments. This paper analyzes how the European AI Act can influence reforms in STEM education from a specific viewpoint. It asserts that generative AI and regulatory frameworks necessitate a revaluation of European pedagogy, assessment, and teacher autonomy. The qualities are interconnected, since evaluation systems influence pedagogical design and the evolving roles of teachers as professionals in technologically intricate and regulated environments. The paper examines policy, educational practice, and disciplinary knowledge to enhance the discourse on making STEM education in Europe pedagogically pertinent, ethically sound, and socially accountable. 2 The European AI Act and education: A new regulatory landscape The EU’s Artificial Intelligence Act is the inaugural comprehensive legal framework for AI governance, categorizing applications based on their potential effects on fundamental rights and societal values. The regulation, while not specific to education, applies to contexts in which AI systems influence teaching, learning, and assessment decisions. AI systems utilized in education for assessing learning outcomes, facilitating student development, or aiding institutional decision-making may be classified as “high-risk” applications. This classification necessitates transparency, human oversight, data integrity, and accountability ( European Union, 2024 ; Veale and Borgesius, 2021 ). In education, regulatory notions directly shape fundamental instructional practices. Transparency concerns arise when AI systems deliver responses or feedback without explaining their underlying processes. In AI-assisted settings, human oversight requires educators to retain authority over instructional and assessment decisions. In STEM education, automated technologies are progressively addressing challenges, elucidating concepts, and evaluating student performance. The AI Act necessitates a balance between human educators and technological systems to ensure that educational processes remain interpretable, accountable, and ethical ( Holmes et al., 2019 ; Williamson and Eynon, 2020 ). Kotsis and Stylos 10.3389/feduc.2026.1845045 Frontiers in Education 02 frontiersin.org In practical terms, human oversight can be documented through simple and proportionate classroom procedures rather than extensive bureaucratic reporting. For example, teachers may keep brief records of the AI tools used in a lesson, the purpose for which they were introduced, and the instructional decisions that remained under teacher control. In assessment contexts, students can submit short AI-use declarations identifying the prompts used, the outputs consulted, and the modifications made. Teachers may also include a rubric criterion related to responsible AI use, transparency, and justification of final answers. Such low-burden documentation supports accountability while preserving teachers’ professional autonomy and avoiding unnecessary administrative overload. Simultaneously, the regulatory framework imposes more challenges for educators and practitioners. There is increasing need to innovate and incorporate AI tools to enhance learning and efficiency. Regulatory constraints may hinder technology deployment or necessitate more assessment and documentation. This concurrent necessity compels educators to reconcile pedagogy, legislation, and ethics. Recent research indicates that numerous educators perceive themselves as inadequately equipped to critically assess AI systems or comprehend regulatory norms, highlighting a disparity between policy expectations and professional capability ( Redecker and Punie, 2017 ; Zawacki-Richter et al., 2019 ). The AI Act signifies a transformation in the paradigm of digital society education. The legislation underscores the moral dimensions of technology, prioritizing human agency, justice, and trust rather than perceiving it as neutral. It reinterprets educational innovation as both technologically informed and socially responsible. AI integration significantly influences STEM education, shaping future citizens’ understanding and utilization of intelligent technologies. Consequently, the AI Act can serve as a catalyst for reimagining European educational design and practice, rather than merely as a limitation. 3 Reconfiguring pedagogy in AI-rich STEM classrooms Generative artificial intelligence in STEM classrooms transcends mere digital tools. These AI systems elucidate, resolve problems, and emulate structured reasoning, in contrast to earlier technologies that facilitated information access and communication. This capability contests conventional educational paradigms that perceive information as a transfer from instructor to student and promotes a distributed learning process that includes both human and non-human agents ( Luckin et al., 2016 ). In STEM education, where understanding is closely linked to genuine disciplinary procedures, AI systems capable of performing complex cognitive tasks present fundamental issues with learning. Physics students typically acquire knowledge through problem- solving, modelling, and iterative reasoning. Nonetheless, when AI technologies can produce complete solutions or offer detailed feedback, students may interact with knowledge representations without thoroughly internalizing the concepts. Cognitive offloading may lead to surface learning that separates procedural fluency from conceptual understanding ( Chiu, 2023 ; Kasneci et al., 2023 ). At the same time, generative AI provides instructional opportunities that align with contemporary learning theories. Social constructivists perceive AI as a cognitive collaborator that promotes inquiry, supports cognitive development, and provides immediate feedback to individual learners. This endorses innovative views of learning environments as flexible and interactive systems in which students actively construct knowledge through debate and exploration ( Luckin, 2016 ). The educator transforms into a learning architect, orchestrating interactions among students, content, and artificial intelligence systems. Artificial intelligence in education is not neutral and must be understood within the legislative framework of the European AI Act. The Act’s focus on transparency and human oversight influences the implementation of AI in classrooms. Pedagogical methods that employ opaque AI-generated outputs without thorough examination may breach principles of explainability and accountability. Consequently, educators must impart “critical AI literacy,” enabling pupils to analyse, understand, and evaluate AI-generated content. This necessitates an epistemic awareness that integrates disciplinary knowledge with the mechanisms through which intelligent systems generate and mediate knowledge ( Kasneci et al., 2023 ; Williamson and Eynon, 2020 ). The pedagogical implications of generative AI vary significantly across STEM disciplines because each field relies on distinct forms of inquiry, reasoning, and knowledge construction. In mathematics education, AI systems can generate symbolic solutions, proofs, and procedural explanations, potentially supporting individualized learning while simultaneously reducing students’ engagement in mathematical reasoning if outputs are accepted uncritically. In chemistry education, AI-supported laboratory simulations and molecular visualization tools may strengthen conceptual understanding of abstract phenomena, although excessive reliance on AI- generated interpretations may weaken experimental reasoning and scientific judgment. In biology education, AI technologies increasingly assist with data interpretation and bioinformatics tasks, requiring students to critically evaluate the reliability and validity of AI-generated analyses. Engineering and technology education present additional challenges because design-oriented learning depends heavily on iterative reasoning, prototyping, and epistemic agency, which may be constrained when AI systems overly structure problem-solving processes ( Gao et al., 2026 ). These disciplinary differences indicate that AI integration in STEM education cannot be approached through generalized digital pedagogy alone. Instead, STEM educators require domain-specific AI literacy frameworks and pedagogical models that preserve inquiry, experimentation, and evidence-based reasoning while enabling students to critically engage with AI- supported knowledge production ( Huwer et al., 2025 ; Leon et al., 2025 ). Artificial intelligence in education raises issues of equity and accessibility. AI tools can customize education and enhance various learners’ capabilities; however, access to technology and digital proficiency may exacerbate inequality. In the absence of meticulous design and institutional backing, technology- enhanced learning environments may reinforce or intensify socioeconomic disparities, particularly in STEM fields ( Zawacki- Kotsis and Stylos 10.3389/feduc.2026.1845045 Frontiers in Education 03 frontiersin.org Richter et al. 2019 ). This subject is closely associated with European inclusion and digital divide policy objectives, emphasizing the necessity for innovative and equitable educational strategies. These concerns should also be situated within broader European policy and funding mechanisms that aim to reduce digital inequalities. Initiatives such as the Digital Education Action Plan, Erasmus+, Horizon Europe, and the European Social Fund Plus can support infrastructure development, teacher professional learning, inclusive digital resources, and research on responsible AI use in education. In the context of the AI Act, such mechanisms are important because regulatory compliance alone cannot guarantee equitable implementation. Schools and teachers require material resources, institutional guidance, and professional development opportunities to ensure that AI-enhanced STEM education does not privilege learners with greater technological access or stronger digital capital. Equity therefore needs to be addressed as both a pedagogical and policy-level responsibility. Reconfiguring pedagogy in AI-enhanced STEM classrooms necessitates transcending the question of whether to employ AI and instead focusing on its impact on learning settings. This necessitates perceiving AI as an epistemic agent that shapes how knowledge is accessed, produced, and validated. Consequently, educators must effectively incorporate AI and provide learning environments that emphasize human knowledge, critical thinking, and disciplinary methodologies. This reconfiguration aligns with the European AI Act’s objective of ensuring that technology advancement upholds human agency and educational integrity. 4 Assessment under pressure: validity, authenticity, and AI Generative artificial intelligence has disrupted STEM assessment practices, raising questions about validity, authorship, and evidence in education. Conventional assessment methods presume that written responses, problem-solving tasks, and standardized tests can effectively evaluate students’ cognitive processes. Nonetheless, AI systems capable of producing high- quality solutions, explanations, and complete assignments undermine this premise, complicating the identification of authentic student work ( Cotton, 2023 ; Kasneci et al., 2023 ). The validity of assessments in AI-enhanced contexts, defined as the extent to which they measure students’ understanding, is crucial. If students utilize AI technology to complete assignments intended to demonstrate their understanding, the validity, defined as the degree to which an assessment measures what it is intended to measure, is undermined. This is essential in STEM disciplines, where problem-solving and analytical reasoning represent core indicators of understanding. When AI systems replicate these processes, the outcome no longer reflects the learner’s own cognitive engagement. This indicates that conventional assessment formats may yield quantifiable results without consistently representing knowledge and proficiency ( Perkins et al., 2024 ). Validity is associated with authorship and academic integrity. AI-generated content blurs the boundaries between human and machine authorship, complicating the establishment and enforcement of originality standards. Institutions have revised policies and explored detection methods; however, current research indicates that these approaches are often inadequate as AI-generated outputs increasingly resemble human-produced work ( Cotton et al., 2023 ). A narrow approach may overlook AI’s educational potential and create tension between students and institutions. Evaluation methodologies must evolve to incorporate AI rather than disregard it. Many scholars have advocated for more authentic and process-focused assessment. Authentic assessment prioritizes complex, contextually rich, and real-world tasks, reducing the likelihood that AI fully substitutes human expertise. STEM education may encompass open-ended inquiries, collaborative projects, oral assessments, and iterative design tasks that promote students to articulate their reasoning and sustain curiosity. These approaches reinforce contemporary views of learning as a process rather than a product, providing stronger evidence of students’ conceptual understanding and epistemic engagement ( Perkins, 2024 ). For example, in an AI-assisted physics assessment, students could be asked to investigate the energy transformations of a pendulum and produce a short scientific report. Generative AI may be permitted for specific purposes, such as suggesting initial hypotheses, identifying possible sources of error, or providing feedback on the clarity of explanations. However, students would be required to submit an AI-use statement documenting the prompts used, the AI-generated suggestions considered, the revisions made, and the reasons for accepting or rejecting particular outputs. The assessment rubric would therefore include criteria such as: accuracy of physics concepts, quality of experimental reasoning, transparency of AI use, evidence of independent interpretation, and ability to justify conclusions orally or in writing. In this model, authorship is not treated simply as the absence of AI use, but as the student’s demonstrated responsibility for the reasoning, decisions, and final scientific claims. Similarly, in an AI-assisted chemistry assessment, students could be asked to investigate a chemical reaction process — such as the factors affecting the rate of a decomposition reaction — and produce a laboratory report. Generative AI may be permitted to support specific steps, such as generating initial predictions, proposing experimental variables, or suggesting safety considerations. However, students would be required to submit an AI-use statement documenting the prompts used, the AI outputs consulted, the modifications made, and the scientific reasoning applied in evaluating those outputs. The assessment rubric would therefore include criteria such as: accuracy of chemistry concepts, quality of experimental design and data interpretation, transparency of AI use, evidence of independent scientific judgment, and ability to critically evaluate AI- generated explanations against empirical evidence. As in other STEM disciplines, authorship is understood not as the absence of AI use, but as the student’s demonstrated responsibility for the scientific reasoning, methodological decisions, and final claims. Similarly, in an AI-assisted biology assessment, students could be asked to investigate a genetics or ecology problem — such as analysing population data to identify factors affecting biodiversity — and produce a scientific report. Generative AI may Kotsis and Stylos 10.3389/feduc.2026.1845045 Frontiers in Education 04 frontiersin.org be permitted to support tasks such as generating initial hypotheses, identifying relevant variables, or interpreting statistical patterns in ecological data. However, students would be required to submit an AI-use statement documenting the prompts used, the outputs consulted, the modifications made, and the biological reasoning applied in evaluating those outputs. The assessment rubric would therefore include criteria such as: accuracy of biological concepts, quality of data analysis and interpretation, transparency of AI use, evidence of independent scientific judgment, and ability to critically evaluate AI-generated explanations against empirical evidence. As in other STEM disciplines, authorship is understood not as the absence of AI use, but as the student’s demonstrated responsibility for the scientific reasoning, methodological decisions, and final claims. In mathematics education, an AI-assisted assessment could ask students to investigate a real-world modelling problem — such as analysing the growth of a population using exponential functions — and produce a written justification of their reasoning. Generative AI may be permitted to support specific steps, such as generating alternative solution strategies, verifying symbolic procedures, or proposing graphical representations. However, students would be required to submit an AI-use statement documenting the prompts used, the AI-generated strategies considered, the modifications made, and the mathematical reasoning applied in selecting or rejecting particular approaches. The assessment rubric would therefore include criteria such as: accuracy of mathematical concepts, quality of reasoning and justification, transparency of AI use, evidence of independent interpretation, and ability to explain conclusions in writing or orally. As in other STEM disciplines, authorship is understood not as the absence of AI use, but as the student’s demonstrated responsibility for the mathematical reasoning, decisions, and final claims. Requirements concerning transparency, fairness, and accountability under the European AI Act directly shape assessment practices. AI-driven grading, feedback, and decision- making systems must be transparent and overseen by humans. It necessitates the implementation of pedagogically, legally, and ethically robust assessment systems, which influence the design of assessment practices. Automated evaluation systems must be scrutinized for bias and include meaningful human input in decision-making. The AI Act underscores the necessity for a thorough examination of the utilization of AI by students and institutions in evaluative contexts ( European Commission, 2024 ). In practice, the principles of the European AI Act The influence of AI on assessment may reshape STEM assessment through greater emphasis on transparency, documentation, and human oversight in AI-assisted learning environments. For example, students may be required to submit short AI-use statements identifying the prompts used, the AI-generated suggestions consulted, and the modifications made during problem-solving or scientific inquiry tasks. In mathematics and engineering education, assessment may increasingly focus on students’ justification of reasoning processes rather than only final answers produced with AI assistance. Similarly, in science laboratory contexts, students may be asked to critically evaluate AI-generated interpretations against experimental evidence and disciplinary principles. Such approaches align with emerging research emphasizing that AI integration in STEM education requires domain-specific pedagogical frameworks capable of preserving inquiry, experimentation, and evidence-based reasoning while ensuring responsible and transparent AI use ( Huwer et al., 2025 ; Leon et al., 2025 ). These developments also reinforce the importance of maintaining meaningful human oversight in educational decision- making. Under the AI Act, teachers remain responsible for interpreting assessment evidence, evaluating conceptual understanding, and ensuring that AI-supported learning processes do not undermine students’ epistemic engagement or scientific reasoning. Consequently, the integration of generative AI into STEM assessment requires not only technological adaptation but also new forms of pedagogical and ethical accountability. The influence of AI on assessment is a chance to reevaluate the objectives and methodologies of evaluation in STEM education. Instructors are increasingly required to develop assessment systems that prioritize understanding, reasoning, and critical engagement with disciplinary content and technological tools, rather than concentrating on efficiency and conformance. This reorientation facilitates advanced cognitive objectives and equips students for interactions with artificial intelligence. There is a need for new assessment models that are robust, meaningful, and aligned with AI-augmented learning in this evolving landscape. 5 Teacher agency in the age of regulation and artificial intelligence The increasing integration of artificial intelligence into educational and legislative frameworks, such as the European AI Act, is reshaping STEM teacher agency. Teacher agency is commonly understood as educators’ capacity to make informed, independent judgments regarding pedagogy, curriculum, and assessment, grounded in professional expertise and contextual understanding ( Biesta et al., 2015 ). In AI-rich and increasingly regulated environments, emerging technological, ethical, and institutional constraints are reshaping agency and redefining the professional responsibilities of educators. The redistribution of decision-making authority between human educators and AI systems is a central element of this transformation. As AI technologies advance in generating educational content, delivering feedback, and recommending pedagogical strategies, educators may shift into facilitators or overseers of algorithmically mediated activities. Recent research suggests that excessive reliance on AI systems may diminish instructors’ professional judgment and instructional creativity ( Holmes et al., 2019 ; Selwyn, 2019 ). At the same time, AI can support teacher autonomy by providing richer data, facilitating differentiated instruction, and enabling more adaptive teaching methods. The objective is to ensure that AI augments teachers’ reflective practice. The European AI Act compounds the problem by imposing requirements for human control, responsibility, and openness. These principles protect users and facilitate ethical AI implementation, although they require educators to bridge the gap between technology systems and pupils. This necessitates that educators utilize AI tools proficiently, critically evaluate their outcomes, comprehend their constraints, and adhere to legal and ethical standards. Professional competence must Kotsis and Stylos 10.3389/feduc.2026.1845045 Frontiers in Education 05 frontiersin.org encompass “AI literacy” and “algorithmic awareness” alongside pedagogical expertise ( Long and Magerko, 2020 ; Williamson and Eynon, 2020 ). In this context, AI legal literacy refers to teachers’ capacity to understand the basic regulatory principles that shape the educational use of AI and to translate them into classroom decisions. This includes recognising when AI tools may influence assessment, feedback, student profiling, or access to learning opportunities; understanding the importance of transparency, accountability, data protection, and human oversight; identifying possible risks of bias or unfair treatment; and documenting how professional judgement remains central in AI-supported teaching and assessment. It also involves the ability to explain to students why AI use must be transparent, how responsibility for final work is established, and how AI-generated outputs should be critically examined rather than passively accepted. Thus, AI legal literacy is not a specialist legal competence, but a practical professional capacity that enables teachers to use AI responsibly within the regulatory and ethical expectations of European education. These evolving roles create tensions between autonomy and regulation. The AI Act underscores human agency by mandating human monitoring for significant choices. However, compliance with legal frameworks may constrain instructors’ capacity to experiment with new technologies or adopt innovative practices, particularly under ambiguous or rigid institutional standards. STEM education, where technological development often outpaces policy, confronts educators with ambiguous and sometimes contradictory demands ( Selwyn, 2019 ). The reconfiguration of teacher agency must be examined within the framework of professional identity and teacher education. As AI increasingly permeates education, educators must be adept technology users and discerning analysts of its educational and societal ramifications. Digital competency frameworks such as DigCompEdu may require expansion to encompass generative AI and regulatory concerns ( Redecker and Punie, 2017 ). In the era of artificial intelligence, teacher autonomy necessitates institutional backing, professional advancement, and alignment between policy and practice, rather than solely individual competence. At the intersection of artificial intelligence and regulation, teacher agency must be reconceptualized as dynamic and contextually dependent. Technology and regulation may constrain teacher agency; yet, new forms of expertise that integrate educational knowledge, ethical reasoning, and technical understanding may reconfigure it. This viewpoint endorses the European AI Act’s objective of having human agents influence the ethical application of AI in education. Educators are central to ensuring that STEM education fosters meaningful learning, analytical reasoning, and informed technological application in this evolving landscape. 6 Toward a conceptual framework for AI-regulated STEM education Prior research indicates that generative artificial intelligence and the European AI Act necessitate a thorough revaluation of STEM education, which can be conceptualized through a framework encompassing policy, pedagogy, assessment, and teacher agency. This framework treats these dimensions as an interconnected system in which changes in one domain influence the others. The AI Act embeds transparency, accountability, and human oversight into the design and use of educational technology ( European Union, 2024 ; Veale and Borgesius, 2021 ). Figure 1 illustrates this framework by showing the dynamic and reciprocal interaction among policy, pedagogy, assessment, and teacher agency in AI-regulated STEM education. These regulatory principles are operationalized through institutional policies and practices that govern STEM curriculum design, assessment, and the integration of AI tools. Consequently, educational institutions interpret and implement regulatory requirements, balancing innovation with compliance. In educational settings, educators and learners engage directly with AI systems, which transform pedagogical interactions. The mediating role of teacher agency involves navigating regulatory constraints and technological opportunities to shape learning experiences ( Priestley et al., 2015 ; Williamson and Eynon, 2020 ). This framework highlights the interdependence between pedagogy and assessment. Pedagogy should emphasize learning, critical engagement, and epistemic comprehension when generative AI disrupts conventional assessment methods. Conversely, assessment practices shape instructional design by redefining meaningful learning objectives. Regulatory requirements for transparent, equitable, and accountable pedagogy and assessment influence these reciprocal relationships. This viewpoint suggests that the incorporation of AI in STEM education entails a systemic transformation involving multiple stakeholders and levels, rather than mere tool implementation. The framework may facilitates the analysis and development of innovative, value-driven, AI-enhanced educational settings by prioritizing policy, practice, and professional autonomy. This framework highlights the interdependence between pedagogy and assessment. Pedagogy should emphasize learning, critical engagement, and epistemic comprehension when generative AI disrupts conventional assessment methods. Conversely, assessment practices shape instructional design by redefining meaningful learning objectives. Regulatory requirements for transparent, equitable, and accountable pedagogy and assessment influence these reciprocal relationships. To make this framework more operational, consider the example of an AI-assisted physics project in which students investigate energy transformation in a pendulum. At the level of policy, the use of generative AI must align with the principles of the European AI Act, particularly transparency, accountability, and human oversight. At the level of pedagogy, students may use AI tools to generate hypotheses, receive formative feedback, or compare alternative explanations, but they are also required to critically evaluate AI-generated outputs against physical principles and experimental evidence. At the level of assessment, the teacher evaluates not only the final product, but also the learning process, including students’ prompts, revisions, interpretations of data, and justification of conclusions. At the level of teacher agency, the teacher acts as a pedagogical mediator who decides when and how AI use is appropriate, documents the role of AI in the learning process, and ensures that students remain responsible for conceptual understanding and scientific reasoning. Similar forms of AI-regulated practice may also emerge across other STEM disciplines. In mathematics education, students may use generative AI tools to compare alternative solution strategies Kotsis and Stylos 10.3389/feduc.2026.1845045 Frontiers in Education 06 frontiersin.org in problem-based STEM tasks or to support mathematical modelling activities connected to real-world engineering challenges ( Crowder et al., 2024 ; Tasarib et al., 2025 ). AI systems may also assist students in identifying mathematical structures or verifying symbolic procedures, while teachers continue evaluating the interpretation, justification, and conceptual understanding underlying mathematical reasoning ( Roberts et al., 2022 ). In chemistry education, generative AI may support context-based STEM activities such as analysing chemical reaction processes, generating molecular visualisations, or assisting students during cheminformatics and 3D-printing projects involving molecular modelling ( Pernaa, 2022 ; Utmeemang and Buaraphan, 2024 ). Similarly, AI-supported laboratory simulations may help students formulate hypotheses and interpret experimental data, although teachers must ensure that learners critically evaluate the scientific validity of AI- generated explanations and maintain engagement with empirical reasoning. In biology education, AI tools may assist students during genetics-focused STEM inquiry activities by supporting data interpretation, critical thinking, and evidence evaluation ( Yaki, 2022 ). AI systems may also support interdisciplinary tasks involving ecology, water pollution, or biological data analysis through spreadsheet-based and adaptive learning environments ( Anwar et al., 2022 ; Kim et al., 2025 ). In such contexts, students must critically examine the reliability and limitations of AI- supported analyses rather than passively accepting generated outputs. Engineering and technology education present additional challenges because design-oriented learning depends heavily on iterative reasoning, prototyping, and epistemic agency. Generative AI systems may support students during engineering design cycles, including microbial fuel cell construction projects or AI-assisted prototyping activities, by generating alternative design ideas and supporting optimization processes ( Tan and Yew-Jin, 2022 ). However, teachers must maintain oversight of problem framing, conceptual understanding, and decision-making throughout the design process to ensure that AI enhances rather than replaces students’ disciplinary reasoning and creative engagement. These examples illustrate that the proposed framework is not limited to physics education but may support diverse forms of AI- enhanced inquiry, modelling, experimentation, reasoning, and assessment across STEM disciplines. In this way, the framework is not limited to a theoretical description of AI-regulated STEM education; rather, it can guide the design of classroom practice in which regulation, pedagogy, assessment, and professional judgment are understood as mutually dependent dimensions. 7 Implications and future directions This paper has significant implications for STEM education stakeholders across Europe, particularly in light of the legal mandates of the European AI Act. The literature suggests the FIGURE 1 Conceptual framework for AI-regulated STEM education, showing the dynamic interaction among policy, pedagogy, assessment, and teacher agency, illustrated through the example of an AI-assisted physics project. Kotsis and Stylos 10.3389/feduc.2026.1845045 Frontiers in Education 07 frontiersin.org necessity for educators to develop new professional competencies that integrate pedagogical expertise with AI literacy and ethical reasoning. Educators must now employ AI tools proficiently, critically evaluate their outputs, comprehend their limitations, and guide students in their ethical usage. This necessitates continuous professional development that transcends technical training to foster reflective and informed teaching ( Long and Magerko, 2020 ). Teacher education programs must reevaluate their curricula to integrate AI. DigCompEdu provides a solid foundation; nonetheless, generative AI presents epistemological and ethical challenges that require attention. Incorporate AI capabilities into subject-specific teaching methods to equip future STEM educators to effectively integrate AI into their disciplines, emphasizing conceptual understanding and critical thinking ( Redecker and Punie, 2017 ). In Greece and across Europe, this entails aligning teacher education with legislative frameworks such as the AI Act and digital education strategies. This analysis highlights the necessity for policymakers to assist schools in reconciling innovation with compliance. Regulatory requirements must be explicit, accompanied by resources and institutional support mechanisms to promote educational experimentation and the responsible use of AI. STEM education research should progress beyond transient evaluations of tool efficacy to investigate the long-term impacts of AI on learning methodologies, assessment practices, and professional obligations. To establish sustainable and equitable AI-enhanced educational systems, such research must integrate pedagogical, technological, and policy perspectives. This paper has several limitations that should be acknowledged. First, the study is conceptual in nature and does not provide empirical classroom-based evidence regarding the implementation of AI-regulated STEM education practices. The arguments presented are grounded in recent theoretical and empirical literature on generative AI, STEM pedagogy, and AI governance rather than direct observational or intervention data. Second, although the manuscript incorporates examples from multiple STEM disciplines, some illustrations remain more developed in physics education due to the authors’ disciplinary background and the existing concentration of AI-related STEM education research in specific scientific fields. Third, the rapid evolution of generative AI technologies and the continuing interpretation of the European AI Act mean that educational implications may vary across institutional, disciplinary, and national contexts. Future research should therefore investigate how AI regulation influences pedagogy, assessment, teacher agency, and epistemic practices across diverse STEM disciplines and educational systems. In particular, empirical studies examining authentic classroom implementation, disciplinary differences, and students’ critical engagement with AI-generated knowledge would contribute significantly to the development of sustainable and pedagogically meaningful AI-enhanced STEM education frameworks. 8 Conclusion Generative artificial intelligence and the European AI Act signify a pivotal moment for STEM education in Europe. This paper contends that AI necessitates a revaluation of pedagogy, evaluation, and educator autonomy, rather than merely the introduction of new instruments. These dimensions are interconnected, and their alteration signifies the production, validation, and implementation of educational knowledge. The AI Act serves as an enabling framework that fosters educational concepts such as transparency, accountability, and human oversight. Regulation and innovation should be understood as interdependent forces in responsible educational practices. Educators and institutions must navigate this evolving landscape while maintaining the primacy of human understanding and leveraging AI to enhance learning. The ability to integrate technological, pedagogical, and ethical considerations into cohesive and sustainable frameworks will shape STEM education in Europe. By actively participating in this process, educators can ensure relevant, equitable, and critically informed educational experiences using AI. Data availability statement The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author. Author contributions KK: Writing – review & editing, Writing – original draft. GS: Writing – original draft, Writing – review & editing. Funding The author(s) declared that financial support was not received for this work and/or its publication. Conflict of interest The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The author KK declared that they were an editorial board member of Frontiers, at the time of submission. This had no impact on the peer review process and the final decision. Generative AI statement The author(s) declared that generative AI was not used in the creation of this manuscript. Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us. Kotsis and Stylos 10.3389/feduc.2026.1845045 Frontiers in Education 08 frontiersin.org Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher. References Anwar, S., Menekse, M., Guzey, S., and Bryan, L. A. (2022). The effectiveness of an integrated STEM curriculum unit on middle school students’ life science learning. J. Res. Sci. Teach. 59 (7), 1204 – 1234. doi: 10.1002/tea.21756 Biesta, G., Priestley, M., and Robinson, S. (2015). The role of beliefs in teacher agency. Teach. Teach. 21 (6), 624 – 640. doi: 10.1080/13540602.2015.1044325 Chen, S., and Cheung, A. C. K. (2025). Effect of generative artificial intelligence on university students learning outcomes: a systematic review and meta-analysis. Educ. Res. Rev. 49, 100737. doi: 10.1016/j.edurev.2025.100737 Chiu, T. K. F. (2023). The impact of generative AI (GenAI) on practices, policies and research direction in education: a case of ChatGPT and midjourney. Interact. Learn. Environ. 32 (10), 6187 – 6203. doi: 10.1080/10494820.2023.2253861 Cotton, D. R. E., Cotton, P. A., and Shipway, J. R. (2023). Chatting and cheating: ensuring academic integrity in the era of ChatGPT. Innov. Educ. Teach. Int. 61 (2), 228 – 239. doi: 10.1080/14703297.2023.2190148 Crowder, A., Flye, R., and Umland, K. (2024). “Deepening mathematical understanding through a problem-based integrated STEM curriculum,” in The Contribution of Mathematics to School STEM Education: Current Understandings , eds. J. Anderson, and K. Makar (Singapore: Springer Nature), 135 – 149. doi: 10. 1007/978-981-97-2728-5_8 European Union (2024). Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 Laying Down Harmonised Rules on Artificial Intelligence and Amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence Act) . Luxembourg: Publications Official Journal of the European Union. Available online at: https://eur-lex.europa.eu/eli/reg/2024/1689/oj (Accessed March 26, 2026). Fan, L., Deng, K., and Liu, F. (2025). Educational impacts of generative artificial intelligence on learning and performance of engineering students in China. Sci. Rep. 15, 26521. doi: 10.1038/s41598-025-06930-w Gao, L., Jong, M. S.-Y., Chai, C. S., and Li, K. (2026). STEM Education: understanding secondary students’ epistemic cognition in the design process with the support of a personalized multi-agent system. Comput. Educ. 242, 105504. doi: 10.1016/j.compedu.2025.105504 Holmes, W., Bialik, M., and Fadel, C. (2019). Artificial Intelligence in Education: Promises and Implications for Teaching and Learning . Boston: Center for Curriculum Redesign. Available online at: https://curriculumredesign.org/our- work/artificial-intelligence-in-education/ (Accessed March 1, 2026). Huwer, J., Thyssen, C., Becker, S., and Thoms, L.-J. (2025). Competencies for teaching with and about artificial intelligence in the natural sciences — DiKoLAN AI. Comput. Educ. Open 9, 100303. doi: 10.1016/j.caeo.2025.100303 Kasneci, E., Seßler, K., Küchemann, S., Bannert, M., Dementieva, D., Fischer, F., et al. (2023). ChatGPT for good? On opportunities and challenges of large language models for education. Learn. Individ. Differ. 103, 102274. doi: 10.1016/j.lindif.2023.102274 Kim, J., Park, W., Erskine, A., and Wang, H.-H. (2025). “Evaluation of a 9th-grade integrated STEM curriculum connecting biology, data analysis with excel, and problem-solving (evaluation),” in ASEE Annual Conference and Exposition, Conference Proceedings . doi: 10.18260/1-2-56458 Leon, C., Lipuma, J., and Oviedo-Torres, X. (2025). Artificial intelligence in STEM education: a transdisciplinary framework for engagement and innovation. Front. Educ. 10, 1619888. doi: 10.3389/feduc.2025.1619888 Li, Y., Tolosa, L., Rivas, F., and Marquez, R. (2026). Translating UNESCO artificial intelligence guidelines to chemical education and its intersection with sustainable development goals. J. Chem. Educ. 103 (3), 1135 – 1144. doi: 10.1021/acs.jchemed.5c00819 Long, D., and Magerko, B. (2020). “What is AI literacy? Competencies and design considerations,” in in Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems, Association for Computing Machinery, Article 2 . doi: 10. 1145/3313831.3376727 Luckin, R., Holmes, W., Griffiths, M., and Forcier, L. B. (2016). Intelligence Unleashed: An Argument for AI in Education . London: Pearson. Perkins, M., Furze, L., Roe, J., and MacVaugh, J. (2024). The AI assessment scale (AIAS): a framework for ethical integration of generative AI in educational assessment. J. Univ. Teach. Learn. Pract. 21 (6), doi: 10.53761/q3azde36 Pernaa, J. (2022). Possibilities and challenges of using educational cheminformatics for STEM education: a SWOT analysis of a molecular visualization engineering project. J. Chem. Educ. 99 (3), 1190 – 1200. doi: 10.1021/acs.jchemed. 1c00683 Priestley, M., Biesta, G., and Robinson, S. (2015). Teacher Agency: An Ecological Approach . London/New York: Bloomsbury. doi: 10.5040/9781474219426 Redecker, C., and Punie, Y. (2017). European Framework for the Digital Competence of Educators: DigCompEdu (JRC107466) . Luxembourg: Publications Office of the European Union. Available online at: https://publications.jrc.ec.europa.eu/ repository/handle/JRC107466 (Accessed February 9, 2026). Roberts, T., Maiorca, C., Jackson, C., and Mohr-Schroeder, M. (2022). Integrated STEM as problem-solving practices. Investig. Math Learn. 14 (1), 1 – 13. doi: 10. 1080/19477503.2021.2024721 Selwyn, N. (2019). Should Robots Replace Teachers? AI and the Future of Education . Cambridge, United Kingdom: Polity Press. Tan, T. T. M., and Lee, Y.-J. (2022). Building improvised microbial fuel cells: a model integrated STEM curriculum for middle-school learners in Singapore. Educ. Sci. 12 (6), 417. doi: 10.3390/educsci12060417 Tasarib, A., Rosli, R., and Rambely, A. S. (2025). Exploring the implementation of STEM education through mathematical modelling activities in schools: a bibliometric analysis. LUMAT 12 (4) 17. doi: 10.31129/LUMAT.12.4.2444 Utmeemang, R., and Buaraphan, K. (2024). Effects of small-scale chemistry STEM integrated with local contexts for enhancing grade 11 students’ learning achievement and learning and innovation skills. Eclética Química 49, e1561. doi: 10.26850/1678-4618.eq.v49.2024.e1561 Veale, M., and Borgesius, F. Z. (2021). Demystifying the draft EU artificial intelligence act. Comput. Law Rev. Int. 22 (4), 97 – 112. doi: 10.9785/cri-2021-220402 Vieriu, A. (2025). The impact of artificial intelligence on students’ learning processes and academic performance. Educ. Sci. 15 (3), 343. doi: 10.3390/ educsci15030343 Williamson, B., and Eynon, R. (2020). Historical threads, missing links, and future directions in AI in education. Learn. Media Technol. 45 (3), 223 – 235. doi: 10.1080/ 17439884.2020.1798995 Yaki, A. A. (2022). Fostering critical thinking skills using integrated STEM approach among secondary school biology students. Eur. J. STEM Educ. 7 (1), 6. doi: 10.20897/ ejsteme/12481 Zawacki-Richter, O., Marín, V. I., Bond, M., and Gouverneur, F. (2019). Systematic review of research on artificial intelligence applications in higher education. Int. J. Educ. Technol. High. Educ. 16, 39. doi: 10.1186/s41239-019-0171-0 Zhang, H., Cai, B., Li, L., and Zhang, R. (2025). “The negative impact of AI-enhanced interdisciplinary STEM curriculum on cognitive skills,” in International Conference on Artificial Intelligence and Education, ICAIE 2025 . doi: 10.1109/ICAIE64856. 2025.11158270 Kotsis and Stylos 10.3389/feduc.2026.1845045 Frontiers in Education 09 frontiersin.org

## Cited law provisions (1)

### GDPR — gdpr-art-2-en

Material scope

---
Generated by overview.legal · https://overview.legal/posts/53829 · 2026-08-22