Artificial intelligence can be developed in one country, trained on data gathered across many others and offered worldwide within days. The people affected by an AI system may therefore live far from the company, computer infrastructure and regulator responsible for it. No world government controls this system, and no single international AI law applies everywhere.
AI governance is instead a layered arrangement. National and regional governments make enforceable laws. Courts and regulators interpret them. International organizations build common principles and scientific evidence. Technical bodies create standards, while companies decide how many safeguards work in practice. Understanding who can do what is the first step toward judging whether this patchwork is adequate.
The UN has built a forum, not a global regulator
UN members created two new mechanisms through General Assembly Resolution 79/325. The Independent International Scientific Panel brings together 40 experts appointed for three-year terms. It produces evidence-based assessments of AI opportunities, risks and impacts, making the same information available to governments with very different technical resources.
The panel is not a regulator. It cannot prohibit an AI product, fine a company or order a government to change its policy. Its preliminary July 2026 report warned that safeguards are failing to keep pace with capabilities and that existing governance tools are fragmented, often voluntary and rarely measured for real-world effectiveness.
The second mechanism is the Global Dialogue on AI Governance. Its first session was held in Geneva on July 6 and 7, 2026, and its next is planned for New York in May 2027. The dialogue gives every UN member a seat alongside researchers, companies and civil society, but it is primarily a venue for discussion and coordination. It does not issue binding global rules.
This division is deliberate. The panel supplies a shared scientific foundation; the dialogue turns evidence into political debate. Together they may help countries agree on problems, vocabulary and possible action without pretending that a UN meeting can instantly replace domestic law.
Enforceable rules remain national and regional
The European Union has adopted the most developed cross-sector regulatory model. Its AI Act classifies uses by risk, bans a narrow group of practices and imposes duties on providers of high-risk and general-purpose systems. Because access to the EU market is valuable, the law can influence companies based elsewhere, much as European privacy rules affected global business practices.
Other governments emphasize different priorities. The United States combines sector-specific law, state rules, procurement requirements, enforcement under existing consumer and civil-rights law, and voluntary commitments. China uses national regulations for algorithms, synthetic media and generative services alongside security review, content duties and industrial policy. In July 2026, China also released an international AI ethics plan calling for lifecycle governance, tiered risk controls and cooperation with developing countries.
These approaches overlap but are not interchangeable. A system may comply with one jurisdiction and still violate rules elsewhere. Governments also differ over privacy, state access to data, competition, copyright, military uses, public speech and how much risk is acceptable in exchange for innovation.
Standards fill gaps without becoming laws
The OECD AI Principles, technical standards from ISO and IEC, and risk frameworks from national institutes help organizations turn broad ideas into testing and management practices. Standards can define vocabulary, documentation, security controls or ways to evaluate bias and reliability. Governments may incorporate them into procurement rules or legislation, making an initially voluntary standard more consequential.
Standards still have limits. Passing a technical test does not answer every democratic question about whether facial recognition should be used in public, whether workers should be monitored or who should bear responsibility when an autonomous system causes harm. Standard-setting can also favor wealthy participants able to attend meetings, run evaluations and shape technical definitions.
Companies govern through design and access
A small number of firms control leading models, specialized chips, cloud infrastructure and large datasets. Their choices about safety testing, acceptable use, release methods and access restrictions can affect millions of people before legislatures act. Open-source developers and downstream companies add another layer because a general model can be modified for purposes its original developer never anticipated.
Company rules can react quickly, but they are not a substitute for public accountability. A business may change its policy, withdraw from a market or define safety differently from affected communities. Independent researchers also need enough access to test claims without exposing private data, trade secrets or dangerous capabilities.
Why countries cannot solve the problem alone
Some risks travel easily across borders. AI-generated fraud can target victims worldwide. A model vulnerability can be copied, and automated disinformation can move between languages and elections. Competition among countries can also discourage tougher rules if leaders fear companies and investment will move elsewhere.
Other questions should remain local. Rules for schools, courts, welfare decisions, policing and workplace monitoring depend on national law and public expectations. International cooperation works best when it supports common safety baselines and information-sharing while leaving legitimate policy choices to democratic institutions.
What a workable system could look like
A realistic global system is more likely to resemble aviation or finance than a single planetary AI agency. Countries would retain legal authority, recognize some compatible standards, report major incidents and cooperate on cross-border enforcement. Scientific bodies would update the evidence, and international forums would expose gaps affecting countries with limited regulatory capacity.
The central challenge is accountability across the chain. Developers know the model, cloud providers control computing access, deployers choose the setting, and governments define legal rights. Effective governance must assign duties to each actor instead of allowing everyone to blame someone else.
What to watch
The UN panel plans thematic briefs and a full annual report before the second Global Dialogue in May 2027. Watch whether countries turn the Geneva discussion into shared testing methods, incident reporting or capacity-building, and whether those commitments include independent measurement.
Also watch implementation of the EU AI Act, new Chinese measures and national court cases involving copyright, discrimination and liability. The decisive question is not whether the world creates one AI regulator. It is whether the institutions already claiming authority can make their rules compatible, enforceable and understandable to the people affected.
Sources
- UN Scientific Panel, global: preliminary assessment and future work.
- UN panel FAQ, global: mandate, membership and limits.
- UN Global Dialogue, global: legal basis and meeting schedule.
- European Commission, European Union: AI Act overview.
- OECD AI Principles, international: common governance principles.
- State Council of China, China: 2026 international ethics plan.
How Much Did You Understand?
Check whether you can distinguish the institutions that study, discuss, standardize and legally regulate AI.
This explainer was prepared with AI assistance and reviewed by a human editor. NewsBridges checked English-, Chinese- and Spanish-language official material.