What is the Global Digital Compact for AI governance?
Global AI regulation
The Global Digital Compact is a 2024 UN political agreement on digital cooperation whose AI section sets a shared framework for governing AI in the non-military domain. Adopted as Annex I to the Pact for the Future, it commits states and stakeholders to a risk-based, inclusive approach grounded in human rights, with transparency, accountability, human oversight, standards cooperation, interoperable governance approaches and capacity building. It guides policy and coordination, but it is not a treaty or a directly enforceable global AI law.
What this means
The Compact is broader than AI. It is a UN framework for digital cooperation across connectivity, digital inclusion, human rights online, data governance and AI. The AI material sits mainly in Objective 5, but it also depends on cross-cutting principles elsewhere in the text, especially international law, inclusion, development and human rights.
For AI, the Compact does not try to write a single detailed global rulebook. It sets shared commitments: assess risks and implications, support compatible governance approaches, improve transparency and human oversight, encourage interoperable standards, and build capacity so developing countries are not left out of AI development and governance.
That makes it different from the separate UN mechanisms later built from it, and different from domestic AI laws. The Compact is the umbrella political instrument. National legislation, sector rules, procurement terms and standards still do most of the binding work.
Why it matters
For organisations, the Compact matters because it shows the direction of travel in international AI governance. Even without direct legal force, it gives governments, public buyers, development institutions, standards bodies and civil society a common language for what responsible AI governance should include: public interest, human rights, transparency, accountability, robust human oversight, and attention to language, culture, labour and environmental effects.
It also sharpens a practical point that is easy to miss in high-income markets: the AI divide is itself a governance issue. The Compact treats access to data, compute, skills, local language resources and participation in rule-making as part of AI governance, not as a side topic. That matters to founders expanding into new markets, buyers running cross-border procurement, advisers shaping internal policy, and policymakers trying to align local rules with international expectations.
How it works
Legal status and scope
The Global Digital Compact was adopted by the UN General Assembly on 22 September 2024 as Annex I to the Pact for the Future. It is an intergovernmentally negotiated UN political instrument, adopted by consensus, and it expressly applies in the non-military domain. That gives the Compact significant political weight and broad reach across the UN system, while keeping it outside the category of treaties that states sign and ratify.
The text also matters for what it does not do. It does not create a comprehensive legal code for AI, and it does not provide a detailed technical definition of AI systems. Instead, it sets objectives, principles, commitments and follow-up steps that governments and other stakeholders are expected to carry forward through national law, regional instruments, standards processes, procurement and international cooperation.
The core AI commitments
Objective 5 is where the main AI commitments sit. The Compact says AI governance should be balanced, inclusive and risk-based, with full and equal representation of all countries, especially developing countries, and meaningful participation by all stakeholders. It anchors AI governance in international law, including international human rights law, and expressly notes the relevance of the UNESCO Recommendation on the Ethics of Artificial Intelligence.
Substantively, the Compact asks states and stakeholders to do four main things. First, improve scientific understanding of AI's future directions, opportunities and risks. Second, support interoperability and compatibility between different AI governance approaches, rather than forcing one global rulebook. Third, build capacity, especially in developing countries, to access, develop, use and govern AI. Fourth, promote transparency, accountability and robust human oversight. The text also says AI should be governed in the public interest, support linguistic and cultural diversity, make room for locally generated data where appropriate for development, and take labour, employment and environmental effects seriously.
The Compact also includes institutional commitments inside the UN. It commits to establishing a multidisciplinary Independent International Scientific Panel on AI with balanced geographic representation, and to initiating a Global Dialogue on AI Governance involving governments and other stakeholders. Those commitments are part of the Compact itself, even though the detailed design and later operation of those mechanisms belong on separate pages.
Standards, assurance and accountability
The Compact is not a technical standard, but it directly addresses the standards layer. It calls on standards development organisations to collaborate on interoperable AI standards that uphold safety, reliability, sustainability and human rights. That matters because many practical governance controls, such as testing, documentation, evaluation, incident handling and auditability, are often carried through standards and assurance frameworks rather than through treaty text.
The AI section should also be read alongside broader commitments elsewhere in the Compact. The wider text calls on digital technology companies and developers to apply human rights due diligence and impact assessments throughout the technology life cycle. It also calls for industry accountability frameworks that define responsibilities, increase transparency and commit to auditable public reports. For an operator, that means the AI chapter is not a standalone ethics note. It sits inside a broader governance frame that expects evidence, oversight and reporting discipline.
Capacity building and the AI divide
One of the Compact's clearest policy choices is that AI governance is not only about controlling risk in advanced markets. It is also about widening access and voice. The text repeatedly links AI governance to the Sustainable Development Goals and to closing digital divides. It encourages international partnerships for education and training, wider access to resources such as open AI models and systems, open training data and compute, and stronger participation by micro, small and medium-sized enterprises in the digital economy.
It also points to concrete building blocks: high-performance computing skills, representative high-quality datasets, affordable compute resources, and local approaches that reflect linguistic and cultural diversity. North-South, South-South and triangular cooperation are explicitly named. The Secretary-General was also asked to develop innovative voluntary financing options for AI capacity-building, which underlines that the Compact treats finance as part of governance architecture, not merely as supporting rhetoric.
Implementation and follow-up
The Compact says implementation should happen within countries and at regional and global level, while respecting national policies, national priorities and applicable legal frameworks. In other words, it is designed to travel through existing institutions rather than replace them. It builds on the World Summit on the Information Society process, the Internet Governance Forum, the WSIS Forum and other UN bodies. It asks the Commission on Science and Technology for Development to consider how it can contribute further to implementation, and it places review roles with wider UN economic, development and human rights machinery.
The follow-up model is partly governmental and partly multi-stakeholder. The Compact invites international and regional organisations, the private sector, academia, the technical community and civil society to endorse it voluntarily and take part in implementation. The UN was asked to make those endorsements public from December 2024, and implementation is being tracked through a public map and portal. The Compact also scheduled a high-level review during the General Assembly's eighty-second session, which the UN timeline places in 2027.
As of 4 June 2026, the Compact remains in implementation and review mode. Separate UN AI follow-up work now exists alongside it, but that should not obscure the basic point: the Compact itself is still the umbrella text that states what Member States agreed to do.
Relationship to domestic law and other instruments
For lawyers, buyers and governance leads, the key operational point is hierarchy. The Compact is an umbrella political instrument. Binding duties still come from domestic and regional law, sector regulation, public procurement terms, contract, court decisions and technical standards adopted into practice. A company cannot rely on the Compact instead of complying with local law.
At the same time, the Compact is more than symbolic. Because it was adopted across the UN, it gives countries with very different legal systems a shared reference point. In practice, it is most useful as a top-level governance frame for cross-border policy alignment, stakeholder engagement and board-level oversight. It can also help organisations explain why their AI governance work needs to cover inclusion, capacity, language, labour and development, not only model safety and headline risk.
Examples
Voluntary endorsement and public signalling. A company, university, technical body or civil society organisation that wants to align publicly can use the UN's voluntary endorsement process. The official explanatory note makes clear that endorsement is of the Compact's vision and principles, not a new legal instrument. That makes endorsement a public governance signal, not a substitute for legal or operational work.
Standards alignment. A standards development organisation or sector consortium can treat paragraph 58 as a policy mandate to prioritise interoperable AI standards around safety, reliability, sustainability and human rights. For organisations that later adopt those standards, the practical workflow is to translate broad principles into documented testing, assurance and reporting controls.
Capacity-building partnerships. A development agency, university network or large platform working in lower-capacity markets can structure AI programmes around the Compact's specific priorities: education and training, access to open models and systems, open training data, compute access, high-performance computing skills, representative datasets and local language support. This is where the Compact most clearly turns AI governance into a participation and capability agenda, not only a restraint agenda.
Common misunderstandings
It is a binding world AI law. No. It is a UN political compact, not a treaty and not a directly enforceable global statute.
It creates a global AI regulator. No. The Compact sets commitments and review steps. It does not itself create a single regulator with power to fine, license or ban companies.
It is only about cutting-edge model risk. No. The text is broader. It covers interoperability, standards, capacity, language and culture, labour, environment and development.
It replaces domestic law. No. The Compact says implementation must respect national policies and applicable legal frameworks. Local law still governs binding duties.
It covers military AI. No. The Compact expressly applies in the non-military domain.
Risks and boundaries
Because the Compact is broad and consensus-based, some terms remain open textured. It does not define AI in technical detail, allocate liability, set sector-specific duties, or resolve hard questions about general-purpose models, compute thresholds, frontier testing or enforcement design. Organisations should not over-read it as if it already answered those questions.
There is also a real implementation gap risk. The Compact depends on domestic lawmaking, regional instruments, standards work, financing and institutional follow-up. Some parts are already under way, such as public endorsements, implementation mapping and separate AI follow-up mechanisms. But as of 4 June 2026, the detailed practice of those later mechanisms is still developing, and their future role remains distinct from the Compact's own text.
The biggest practical mistake is to use the Compact either too weakly or too strongly: too weakly, by dismissing it as mere diplomacy; too strongly, by treating it as a complete compliance framework. It is neither. It is a durable international reference point that still has to be translated into local legal and operational controls. Public endorsement should also not be mistaken for evidence of mature AI governance.
What to do next
Start by classifying the Compact correctly in your governance stack: board-level and policy-level frame, not binding law.
Map your current AI controls against its recurring themes: transparency, accountability, human oversight, human rights due diligence, standards alignment, cultural and linguistic inclusion, labour effects, environmental effects and capacity building.
If you operate across regions, use the Compact as a common baseline for internal policy and supplier due diligence, then layer stricter local rules on top.
If you work with governments, development finance or emerging-market partners, examine whether your approach addresses the AI divide, including local data stewardship, language coverage, skills, compute access and SME participation.
Decide whether public endorsement, standards participation or engagement with UN and regional processes would advance your credibility and learning. Then monitor the UN review cycle and the separate AI mechanisms without confusing them with the Compact itself.
FAQs
Is the Global Digital Compact legally binding?
No. It is a UN political instrument adopted by the General Assembly, not a treaty with its own ratification process and not a directly enforceable global AI law.
When was it adopted, and what is its status now?
It was adopted on 22 September 2024 as Annex I to the Pact for the Future. As of 4 June 2026, it is in an implementation and review phase, with public endorsement, implementation tracking and separate UN AI follow-up work proceeding alongside it.
Does the Compact define AI?
Not in a detailed technical way. It uses the term AI systems, but it does not give the kind of precise definition that many domestic laws or technical instruments use.
Does it create a global AI regulator?
No. It creates commitments, follow-up steps and UN discussion routes. It does not itself establish a single regulator with direct enforcement power over organisations.
Does it apply to companies if states adopted it?
Indirectly, yes. Companies are not the adopting parties, but the Compact expressly calls on private actors to respect human rights, use due diligence, support accountability and take part in implementation and endorsement.
Does endorsement make an organisation compliant?
No. Endorsement signals support for the Compact's vision and principles. It does not prove compliance with domestic law, and it does not by itself show that an organisation has mature AI controls.
Does it cover military AI?
No. The Compact states that it operates in the non-military domain.
How should I use it if I already have to comply with domestic AI law?
Use it as a top-level policy and governance frame. It can help align cross-border practice and board oversight, but local law, sector rules and procurement requirements still govern the binding details.