The gap in AI governance lies in implementation rather than in principles. Since 2021, UNESCO’s Recommendation on the Ethics of Artificial Intelligence has been adopted by many countries, and similar declarations have proliferated. The harder problem is implementation: what does an ethical commitment actually require of a ministry, a regulator or a school system on a Tuesday afternoon? UNESCO announced plans to introduce new governance tools at its Global Forum in Riyadh, scheduled for 14-17 September, in an advance notice published on 9 September.
From principles to practice
The organisation points to work with countries on AI readiness and policy gaps. The challenge it addresses is not agreeing that AI should be responsible but developing the institutions and methods to assess how systems are introduced and used. That means readiness assessments, policy frameworks and practical guidance rather than further statements of values.
Why implementation is hard
Principles are general by design; implementation is specific. A commitment to fairness means little until you specify which systems, which populations, which metrics and which remedies. A commitment to transparency requires deciding what must be published and who verifies it. Each of those decisions is technical, political and contested, which is why the translation from principle to practice is where most governance efforts stall.
What readiness assessment involves
A readiness assessment typically examines a country’s legal framework, institutional capacity, data infrastructure and skills base. Its purpose is to identify where the gaps are, so that investment and policy can be targeted. Done well, it is a diagnostic tool; done poorly, it becomes a checklist that produces a score and no action. The design of the method matters as much as the result.
UNESCO can convene, set norms and provide methods, and it cannot enforce. Its leverage comes from legitimacy and from the willingness of governments to be seen complying. That is a real form of influence, and it is slower and weaker than regulation. Understanding the limits of the instrument helps explain why progress on AI governance is uneven across countries.
What to watch for
This brief is based on an advance notice. It does not confirm that every planned tool has been released or that the forum produced particular outcomes. The useful evidence will be the published tools and, more importantly, details of how governments use them. An announcement of a framework is not the same as its adoption, and adoption is not the same as effect.
The recurring pattern
AI governance has followed a recognisable arc: principles, then frameworks, then the difficult work of enforcement. Each stage is necessary and none is sufficient. UNESCO’s tools belong to the middle stage, where the abstractions of ethics meet the specifics of administration. That is unglamorous work, and it is where the outcome is decided.
A readiness assessment usually examines four areas: the legal and regulatory framework, institutional capacity, data and technical infrastructure, and human skills. The output is a picture of where a country stands and what it would need to build. Done well, it produces a prioritised list of actions; done as a formality, it produces a report that is filed and forgotten. The difference lies in whether the assessment is tied to funding and political commitment.
What other bodies are doing
UNESCO is not alone. The OECD maintains an AI policy observatory, the Council of Europe has negotiated a convention on AI, and national regulators are developing their own frameworks. The result is a crowded field with overlapping initiatives, which creates both momentum and confusion. For a country with limited administrative capacity, navigating several frameworks at once is a genuine burden, and harmonisation is one of the less discussed challenges of AI governance.
The enforcement question
Whatever tools UNESCO releases, enforcement remains national. A framework has force only where a country writes it into law, funds a regulator and provides remedies. International bodies can set expectations and offer methods; they cannot compel. That is the structural limit of the approach, and it explains why the same principles produce very different outcomes in different countries.
Countries engage with international AI frameworks for several reasons: access to expertise and methods, legitimacy for domestic policy, and the desire not to be left out of a fast-moving field. Participation is not the same as compliance, and governments vary in how much they implement. For smaller states with limited technical capacity, the tools on offer can be genuinely useful, providing a template they could not build alone. For larger states with their own regulators, the value is more about coordination than guidance.
What ultimately matters is whether the tools change behaviour. A framework that governments cite but do not act on leaves the underlying risks untouched, and the measure of success is not the number of documents published but the number of systems assessed, revised or refused. That is a harder thing to count, and a more honest one to ask about.
This report draws on UNESCO.
Image: U.S. Government Accountability Office from Washington, DC, United States · Public domain · via Wikimedia Commons.