UNESCO Issues Guidance on Generative AI in Education — September 7, 2023

UNESCO issued its Guidance for generative AI in education and research on 7 September 2023, responding to the rapid adoption of large language models in classrooms, universities, and laboratories. The 96-page document interprets UNESCO’s 2021 Recommendation on the Ethics of Artificial Intelligence for the specific context of education and research systems. It sets out policy recommendations for governments and institutions across five pillars: (1) developing and regulating generative AI strategies aligned with human rights, (2) building human capacity and digital literacy, (3) ensuring quality and inclusive learning environments, (4) advancing responsible research and innovation, and (5) establishing robust governance, procurement, and data protection frameworks. The guidance emphasises that generative AI should augment—not replace—human-led education, and that inclusion, equity, gender equality, cultural diversity, and sustainability must anchor deployment decisions.

UNESCO highlights generative AI’s potential benefits: personalised learning support, automated feedback, language translation, and research assistance. However, it warns of significant risks, including bias, disinformation, erosion of academic integrity, mental health impacts, intellectual property violations, and privacy breaches. The guidance calls for a “human-centred” approach where teachers remain in control, learners develop critical thinking, and institutions establish accountability mechanisms. Governments are urged to craft national policies that articulate vision, goals, and ethical principles for generative AI, referencing UNESCO’s Recommendation and other human-rights instruments. Policies should set clear boundaries for use, define prohibited practices (such as biometric surveillance or manipulation), and align with data protection laws ensuring learners and researchers retain rights to access, correct, and delete their information—functionally equivalent to DSARs.

Governance mandates for policymakers

UNESCO recommends that governments form multi-stakeholder governance structures—including ministries of education, research councils, data protection authorities, civil society, industry, and learner representatives—to oversee generative AI strategies. These bodies should conduct national risk assessments, map existing legal frameworks, and coordinate regulatory updates. UNESCO stresses the need to integrate AI policies with existing digital education strategies, cybersecurity plans, and privacy regulations. Governments should establish national guidelines for procurement, requiring vendors to demonstrate ethical compliance, transparency of model capabilities and limitations, data protection measures, and support for accessibility (language, disability accommodations). They should also support independent evaluation of generative AI tools before large-scale deployment.

For accountability, UNESCO urges governments to mandate public reporting on how generative AI is used in education, including metrics on adoption, outcomes, incidents, and mitigation actions. Governments should empower data protection authorities and education inspectorates to audit institutions, review DSAR handling, and investigate complaints. UNESCO emphasises that regulatory sandboxes can enable innovation while maintaining oversight, but they must include safeguards for informed consent, privacy, and redress.

Implementation roadmap for education systems

The guidance proposes an implementation cycle: assess readiness, plan strategy, pilot responsibly, scale with safeguards, and monitor outcomes. Institutions should begin with a situational analysis covering infrastructure, digital skills, teacher attitudes, data governance, and legal obligations. They should create policies that specify permissible use cases (e.g., lesson planning, formative assessment, tutoring), prohibited uses (automated grading without human review, surveillance), and ethical guidelines for staff and students. Training programmes must prepare teachers to integrate generative AI into pedagogy while fostering critical thinking about AI outputs.

UNESCO advises piloting generative AI in controlled settings with clear objectives, evaluation criteria, and stakeholder consent. Pilots should include impact assessments examining equity, inclusion, accessibility, and data protection. Institutions must establish cross-functional committees—combining academic leaders, IT, legal, privacy, ethics, and student representatives—to oversee pilots and recommend scaling decisions. When scaling, institutions should embed human oversight (teachers review AI-generated feedback), maintain transparency (disclosing AI usage to learners), and provide opt-out options consistent with DSAR principles.

Human capacity and digital literacy

The guidance underscores investment in digital literacy for educators, learners, and researchers. UNESCO calls for updated curricula that teach algorithmic thinking, AI ethics, data protection rights, and DSAR processes. Teachers need professional development on how generative AI works, how to evaluate outputs, manage bias, and uphold academic integrity. Institutions should create resources and communities of practice to share best practices, including case studies on DSAR requests involving AI-generated content or training data.

UNESCO emphasises inclusion: digital literacy initiatives must reach marginalised groups, including girls, rural learners, people with disabilities, indigenous communities, and migrants. Programmes should be offered in multiple languages and accessible formats. Governments should invest in infrastructure (connectivity, devices, assistive technologies) to ensure equitable access to generative AI tools. Without such investment, UNESCO warns that generative AI could widen the digital divide.

Responsible research and innovation

For higher education and research institutions, UNESCO outlines responsibilities for using generative AI in scholarly work. Researchers must ensure transparency about AI-generated content, avoid fabricating citations, and respect intellectual property. Institutions should update academic integrity policies to address AI-assisted writing and code generation, including guidelines on attribution and disclosure. Research ethics committees should require data management plans that document training datasets, consent, anonymisation, and DSAR handling. UNESCO encourages open science practices—sharing code, datasets, and evaluation protocols—while safeguarding sensitive data through privacy-enhancing techniques.

The guidance calls for investment in public-interest AI research, including development of multilingual and culturally inclusive models that serve underrepresented languages. Governments should fund collaborations between academia, industry, and civil society to build open, trustworthy generative AI aligned with UNESCO’s ethical principles. Impact assessments should evaluate environmental sustainability, energy consumption, and life-cycle impacts of AI infrastructure.

Data protection, DSARs, and learner rights

UNESCO explicitly links generative AI adoption to data protection and individual rights. The guidance urges compliance with global privacy norms, including lawful basis for processing, purpose limitation, data minimisation, accuracy, storage limitation, integrity, and confidentiality. Institutions must implement mechanisms for learners and educators to exercise access, correction, deletion, and objection rights—effectively DSARs. Policies should specify response timelines, verification methods, and escalation channels to regulators. UNESCO advises institutions to maintain detailed records of data processing, training datasets, retention schedules, and third-party sharing so DSAR responses are accurate and timely.

The guidance warns against opaque data harvesting and recommends clear consent processes, particularly for minors. It calls for age-appropriate protections, including prohibitions on using generative AI for children under a certain age (aligned with local laws), parental consent mechanisms, and educational content about digital safety. Institutions should implement privacy-by-design controls: encryption, access controls, audit logging, and privacy impact assessments for each new AI tool. UNESCO highlights the risk of biometric and emotion recognition technologies, urging strict regulation or bans in educational contexts.

• Grievance mechanisms. Institutions must establish channels—ombuds offices, ethics hotlines, digital complaint portals—where learners and staff can report harms or request redress. These mechanisms should integrate with DSAR processes and provide transparent resolution timelines.
• Transparency artefacts. UNESCO recommends publishing “AI use statements” describing purpose, data sources, decision logic, human oversight, and contact points. These artefacts support accountability and help individuals prepare DSARs.
• Vendor management. Contracts with EdTech providers should include clauses on data ownership, DSAR support, security standards, incident reporting, and audit rights. Vendors must commit to UNESCO’s ethical principles and provide documentation for compliance reviews.

Procurement and ecosystem governance

UNESCO advises governments and institutions to develop procurement frameworks that evaluate generative AI tools against ethical criteria: transparency, explainability, bias mitigation, accessibility, data protection, and sustainability. Requests for proposals should require vendors to supply risk assessments, DSAR support procedures, localisation options, and evidence of accessibility compliance (WCAG standards). Evaluation committees should include privacy officers and educators to balance innovation with safeguards. Post-procurement, institutions must monitor vendor performance, conduct periodic audits, and enforce remediation for non-compliance.

The guidance encourages international cooperation. UNESCO calls for global repositories of trustworthy AI resources, cross-border research partnerships, and harmonisation of standards. It emphasises the need to support low- and middle-income countries through capacity building, funding, and knowledge sharing so they can develop localised generative AI solutions that respect cultural and linguistic diversity.

Monitoring, evaluation, and continuous improvement

UNESCO recommends establishing indicators to track generative AI’s impact on learning outcomes, equity, teacher workload, and DSAR fulfilment. Institutions should collect qualitative and quantitative data, including surveys on learner trust and case studies of DSAR resolutions. Independent evaluations should assess whether generative AI deployments uphold human rights and avoid discriminatory outcomes. Results should inform policy adjustments and be shared with stakeholders, including parents, unions, academia, and regulators.

Finally, UNESCO underscores adaptability. Generative AI evolves quickly; therefore, policies must include review cycles, scenario planning, and mechanisms to sunset tools that no longer meet ethical standards. Embedding DSAR compliance, transparency, and human-centred design into these cycles ensures that generative AI strengthens, rather than undermines, education and research systems worldwide.