By Dr Adam Edmett, Head Ed Tech Innovation, British Council and Dr Helen Crompton, Executive Director, Research Institute of Digital Innovation in Learning (RIDIL), Professor of Instructional Technology, Old Dominion University
Generative artificial intelligence arrived in classrooms faster than the systems meant to govern it. Today, ministries, school leaders and universities continue to grapple with two questions: what should responsible GenAI governance actually cover, and how can it remain useful as the technology changes?
This blog shares the results of two international Delphi studies at primary/secondary and higher education levels. The Delphi method brings together a panel of experts who respond to repeated rounds of consultation, refining their collective view until consensus emerges. We also consider the relevance of the Socio-Ecological Technology Integration (SETI) framework, and the AI guidelines the British Council has recently developed for teachers. Each of these helps to shape a way of thinking about AI governance that can be applied across a range of contexts.
Eight essential areas
The primary and secondary education study involved 24 experts in 20 countries and six continents. The higher education study involved 35 experts in 22 countries and six continents. Both panels developed an eight-part framework showing what policies and guidelines should cover in each education sector, alongside a six-part process for keeping the frameworks current.
Six areas appeared in both frameworks. These were privacy and data protection, ethical and responsible use, equitable access, academic integrity, human oversight and accountability, and GenAI literacy.
Conversations around AI governance often prioritize reducing plagiarism and cheating, but the frameworks show we must go beyond this. AI governance must also address whose data are exposed, who can access approved tools, whether outputs reproduce bias, who remains responsible for decisions and whether learners and educators can judge AI critically.
These areas work together. Access without GenAI literacy can expose learners to confident misinformation. Human oversight without clear accountability leaves uncertainty about who must intervene to solve issues. Privacy restrictions without approved tools or training may drive teachers and students towards unsanctioned platforms. Leaving out one of these framework areas can weaken the protections and educational value promised by the others.
Look more closely and important differences emerge. In primary and secondary education, data privacy and security were leading concerns. Children require age-appropriate safeguards, limits on data collection, family involvement and compliance with child-protection and privacy law. The framework also includes curriculum integration and governance and review, reflecting the need for coordination across national, regional, district and school levels.
In higher education, the emphasis shifts. Academic integrity was the most frequently cited concern, but universities must do more than define unacceptable use. They also need plans for using AI in teaching, learning and assessment, approved tools, technical support and professional learning that works across disciplines. The two additional areas in the higher education framework are therefore integration strategy and institutional support and infrastructure.
Each study asked whether institutions need policies, guidelines or both. Some 89% of the primary and secondary panel and 88% of the higher education panel members supported combining them. Policy provides firm boundaries, while guidelines help educators respond to changing tools and situations. Neither is enough unless institutions decide what must be governed and provide the support needed to put it into practice.
Both studies also examined something often missing from governance conversations. They asked how policies and guidelines can stay current. The panels proposed a dedicated and representative GenAI group, scheduled reviews, continuing professional development, clear communication, evaluation of outcomes and monitoring of external developments. Governance should not be a document written once and left on a shelf.
SETI: Why the ecosystem matters
The Delphi frameworks set out what AI governance should cover. The Socio-Ecological Technology Integration (SETI) framework helps explain how governance works, or fails, across an education system. SETI places the educator at the centre of several connected layers of influence: classroom practice in the microsystem, institutional culture and policy in the exosystem, and wider social forces in the macrosystem. The important point is that decisions made at one level affect what happens at the others.
A national policy may shape an individual teacher’s decision about whether learners should be allowed to use a chatbot for homework but influence also runs in the other direction. Educators cannot integrate technology effectively if the systems around them fail to provide guidance, training or infrastructure. A national policy without school-level support is just a document, and a classroom guideline without institutional backing leaves teachers carrying risks that should be managed elsewhere. Good governance has to work across the whole system.
From principles to practice: British Council AI guidelines
The British Council’s AI guidelines for teachers offer a useful test of both the Delphi frameworks and the SETI perspective. We developed the guidelines to help English language teachers make responsible choices when using AI. They grew out of research involving 1,348 teachers across 118 countries and territories, many of whom told us they lacked the training and support needed to use AI effectively and safely.
In SETI terms, the guidelines focus on the innermost level: the individual teacher and their classroom. Mapped against the eight themes of the Delphi frameworks, the alignment is close. Data privacy, ethical use and bias, transparency about when AI has been used, human oversight, and accountability all appear in both, and the guidelines encourage teachers to involve learners and guardians in decisions where appropriate. They are also explicit that teachers should follow the policies of their own institutions. That is the hybrid model at classroom level: flexible principles for professional judgement nested within firmer institutional rules.
Two aspects of the guidelines are worth noting. The first is environmental impact. The British Council guidelines ask teachers to consider the energy and resource costs of AI use. Neither Delphi panel identified this as a separate governance theme.
The second concerns the limits of teacher-facing guidance. Issues such as equitable access, procurement, infrastructure and system-wide data governance cannot be solved by individual teachers. They sit with institutions, governments and other bodies with the authority and resources to act. This is a reminder that no single document, aimed at one part of the education system, can provide the whole answer.
The guidelines have also been updated with practical checklists. This reflects the kind of living-document approach both expert panels called for.
What this means for policymakers
Three takeaways stand out.
First, start with what governance needs to cover. Every framework should address privacy, ethics, equity, academic integrity, human oversight and GenAI literacy, then decide which elements require formal policy and which need adaptable guidance.
Second, build the review mechanism before you need it. AI policy and guidance have a short shelf life, so regular review should be part of the governance model from the outset, not added later.
Third, context is crucial. School systems will give greater weight to protection, privacy and alignment across levels of governance. Universities will give more weight to academic integrity, professional discretion and infrastructure.
Low- and middle-income countries face another challenge. Implementation of these frameworks requires adequate capacity and commitment, which are sometimes in short supply given resource constraints and competing priorities. Without investment, existing inequalities could widen: high-resource education systems may develop well-governed AI use while lower-resource systems are left with weak or absent governance. Supporting governments to develop and embed these policies and guidelines should be a priority.
Otherwise, the education systems with the fewest resources to manage AI’s risks will be the ones most exposed to them.
