UNESCO, OECD, and the practical case for cross-border AI ethics frameworks

The European AI Act has dominated AI governance conversation since 2024. That is partly deserved — it is the first comprehensive horizontal AI legislation in a major jurisdiction — but the focus has obscured the substantial influence of two older instruments that continue to shape AI policy and practice worldwide: the OECD AI Principles (2019, updated 2024) and the UNESCO Recommendation on the Ethics of Artificial Intelligence (2021).

For practitioners working across borders, understanding when each framework matters and how they fit together is more useful than treating any single framework as definitive.

The OECD Principles: the institutional foundation

The OECD AI Principles were the first intergovernmental standard on AI. Adopted in May 2019, they predate the EU’s High-Level Expert Group on AI Guidelines, the UNESCO Recommendation, the Council of Europe Framework Convention, and the EU AI Act itself.

Their structure is straightforward: five values-based principles for trustworthy AI (inclusive growth and well-being; human rights and democratic values; transparency and explainability; robustness, security and safety; accountability) and five recommendations to governments (R&D investment; digital ecosystem; enabling policy environment; human capacity and workforce; international cooperation).

The institutional adoption was decisive. Within a month of OECD adoption, the G20 endorsed compatible principles at Osaka. Within a year, the OECD AI Policy Observatory at oecd.ai was publishing comparative national policy tracking that has become the reference resource for international AI policy analysis.

In May 2024, the OECD Council updated the Principles to reflect generative AI, GPAI lifecycle considerations, and broader environmental and societal concerns. The update is largely incremental rather than transformative, but it consolidates the Principles’ role as the institutional baseline that other instruments build upon.

For practitioners, the OECD Principles work best as the broadest common denominator in cross-border conversations. Citing them in policy submissions, ethical impact statements, and corporate AI governance frameworks gives the work an internationally recognised anchor without committing to any single jurisdiction’s specific operational requirements.

The UNESCO Recommendation: the global ethical reach

The UNESCO Recommendation on the Ethics of AI was adopted in November 2021 by all 193 UNESCO member states. The unanimity is striking. No other international instrument on AI carries that breadth.

Where the OECD Principles are economically and procedurally framed — fit for high-income governance discussions — the UNESCO Recommendation is normatively and developmentally framed. It is built around four values (human rights and dignity; environmental and ecosystem flourishing; diversity and inclusiveness; peaceful, just and interconnected societies) and ten principles, alongside eleven policy action areas covering ethical impact assessment, gender, culture, education, communication, economy and labour, health, and others.

Two features make the UNESCO Recommendation distinctive in practice. First, environmental and cultural attention. The Recommendation foregrounds AI’s environmental impact and its role in cultural diversity — dimensions less central to the EU AI Act and the OECD Principles. Second, Global South orientation. UNESCO’s mandate spans the entire UN membership; the Recommendation reflects the perspectives of countries that import rather than build most AI systems. This is materially relevant in cross-border engagements involving Latin America, Africa, South Asia, and parts of the Middle East.

UNESCO has operationalised the Recommendation through two tools: the Ethical Impact Assessment (an instrument for organisations evaluating specific AI deployments against the Recommendation’s principles) and the Readiness Assessment Methodology (a country-level diagnostic helping governments evaluate their institutional readiness for AI). Both have been adopted by multiple national authorities since 2022.

For practitioners, the UNESCO Recommendation is the framework to cite when working outside high-income contexts. Capacity-building programmes in countries implementing the Recommendation, advisory work for NGOs and public bodies in the Global South, and corporate AI policies that need universal-reference legitimacy all benefit from explicit UNESCO alignment.

Where the EU AI Act fits

The EU AI Act is not a competitor to UNESCO and OECD frameworks — it is an operational implementation of compatible principles in a specific legal jurisdiction. EU compliance work covers most of what UNESCO and OECD principles require in terms of technical practice: risk management, transparency, accountability, human oversight, fairness, security.

What the EU AI Act adds, beyond UNESCO and OECD, is enforceability. The AI Act has named regulators, penalty structures, market surveillance mechanisms, and conformity assessment requirements. UNESCO and OECD instruments rely on political and reputational incentives instead.

What the EU AI Act lacks, compared to UNESCO and OECD, is scope of acceptance. Organisations operating outside the EU may face counterparties who do not recognise EU regulatory categories. UNESCO and OECD references work across more contexts.

The Council of Europe Framework Convention: the binding international option

A fourth instrument now occupies an important place in this landscape: the Council of Europe Framework Convention on AI, adopted in May 2024 and opened for signature in Vilnius in September 2024. Signed by the EU, the US, the UK, Israel, and several other states, the Framework Convention is the first binding international treaty on AI.

Its structure is principles-based — human rights, democracy, rule of law — but its legal form is treaty law, meaning ratifying states must implement the principles through national legislation. The Convention is open to non-Council-of-Europe states, positioning it as a potential first global binding instrument.

The Framework Convention is best understood as principles-binding rather than operational-binding. It does not replace the EU AI Act, the UNESCO Recommendation, or the OECD Principles. It anchors them in international law for ratifying parties.

A practical map

Decisions about which framework to cite in a given piece of work depend on jurisdiction, counterparty, and stage.

For EU-market activity, cite the EU AI Act as the operational baseline. Add UNESCO and OECD references when the work has a normative dimension — corporate AI policies, public communications, international stakeholder engagement.

For US-, UK-, and Gulf-market activity, cite the OECD Principles as the institutional baseline. Add EU AI Act references when EU exposure exists. Cite UNESCO references when working with international institutions, NGOs, or universities.

For Global South activity, the UNESCO Recommendation is often the most relevant primary reference, supplemented by OECD where applicable and EU AI Act for exports to the European market.

For international policy work — submissions to UN bodies, multilateral consultations, cross-border standardisation — the UNESCO Recommendation and the OECD Principles together form the strongest reference basis. The Council of Europe Framework Convention is becoming increasingly relevant as ratifications accumulate.

AIPIA’s approach

AIPIA’s Code of Ethics is designed to satisfy all four major frameworks simultaneously. The six binding principles — competence, transparency, fairness, accountability, privacy, societal benefit — map cleanly onto OECD Principles, UNESCO values, Council of Europe foundations, and EU AI Act operational requirements.

For practitioners working internationally, AIPIA membership offers a single coherent professional standard that travels across jurisdictions. The European Digital Credential issued by AIPIA carries the same ethical commitments wherever a member operates.

International AI governance is now genuinely plural. The EU AI Act is the loudest instrument, but it is not the only one and probably not the most universally relevant. Practitioners who treat the frameworks as complementary rather than competitive are better positioned for the work that crosses borders — which is most AI work today.