While the EU spent three years building a comprehensive AI law, China enacted targeted regulations on recommendation algorithms, deepfakes, and generative AI in rapid succession.
Faster? Yes. With different objectives? Absolutely. Understanding China's AI governance requires setting aside the assumption that all AI governance shares the same goals.
China has developed one of the world's most active AI regulatory regimes — not through a single comprehensive law, but through a series of targeted regulations on specific AI applications, built on top of a broader state-centric governance philosophy.
Key regulations include: Algorithm Recommendation Regulations (2022) — requiring transparency, user control over recommendation systems, and prohibitions on addictive recommendation practices targeting minors. Deep Synthesis Regulations (2022) — governing deepfakes and synthetic media, requiring content labeling and real-name verification for providers. Generative AI Regulations (2023) — requiring generative AI services to undergo security assessments before public release, with content restrictions aligned with Chinese law and "core socialist values."
Chinese AI governance is not simply "more regulation" — it reflects a fundamentally different theory of what AI governance is for. Where EU regulation centers on individual fundamental rights and where US frameworks emphasize market efficiency, Chinese governance centers on social stability, state security, and "the healthy development of AI." These are not the same goals, and the governance architecture reflects the difference.
China occupies a uniquely complex position in global AI governance. As a major AI developer and exporter, China shapes norms through its technology — Chinese AI systems operating globally carry implicit governance assumptions about data handling, content moderation, and user privacy. As a rule-setter, China's regulations on AI are among the world's most technically specific and actively enforced. And as a state-capitalist system, China's "private" AI companies operate within a governance context where state interests explicitly shape corporate behavior.
Pick one AI application domain: content moderation, hiring, medical diagnosis, or autonomous vehicles.
Compare how the EU AI Act, US voluntary frameworks, and China's approach would govern that application. For each, identify the underlying governance goal that shapes the specific requirements — don't just describe the rules, explain what they're trying to achieve.
AI governance conversations happen most loudly in Brussels, Washington, and Beijing. But the AI systems being governed often operate most consequentially in contexts where those conversations are not happening.
The people most likely to be affected by a facial recognition system or a credit algorithm deployed in Nairobi or Lagos have had limited voice in shaping the frameworks that govern those systems.
When AI governance discussions happen at global forums — the UN, OECD, G20, ITU — they are dominated by US, EU, and Chinese voices. This reflects where most of the world's frontier AI development happens. It does not reflect where most of the world's population lives, or where many of the most consequential AI deployments are occurring.
AI systems are being deployed at scale in Global South contexts: predictive policing tools in Brazil, credit scoring in Kenya, agricultural advisory systems in India, facial recognition in South Africa. These deployments often use systems developed elsewhere, trained on data from elsewhere, and governed by frameworks developed for different contexts.
Access and Capacity: Most developing countries lack the technical expertise to evaluate AI systems, conduct conformity assessments, or develop their own governance frameworks. Regulatory capacity is not evenly distributed. The same regulatory requirements that are manageable burdens for large US tech companies can be existential barriers for local AI developers in countries without mature AI industries.
Data Representation: AI systems trained primarily on data from the US, Europe, or China may perform significantly worse on populations, languages, and environments not represented in training data. Governance frameworks developed around these systems may not identify or address harms that manifest differently in underrepresented contexts.
Regulatory Colonialism: Some scholars argue that when Global South countries adopt EU or US AI governance frameworks wholesale, they import governance assumptions designed for different economic, social, and political contexts — effectively allowing developed-country regulators to shape AI governance in countries that had no seat at the table where the rules were written.
The AI governance frameworks that will shape global AI norms — the EU AI Act, NIST RMF, China's regulations — were all developed without meaningful participation from most of the world's countries. The population most affected by AI governance decisions is not the population that made those decisions.
Choose a specific AI deployment in a Global South context (predictive policing in Brazil, credit scoring in Kenya, facial recognition in South Africa, agricultural AI in India).
Analyze: (1) What governance framework currently applies? (2) What governance gaps exist? (3) How might the EU AI Act or NIST RMF apply — and what would be problematic about that application?
The OECD has principles. The UN has resolutions. The G7 has a Code of Conduct. The ITU has forums. No international body has binding authority over the AI systems that most affect people's lives.
This is not for lack of trying. It reflects genuine disagreements about what AI governance should accomplish — disagreements that are not bridged by adding more signatories to non-binding documents.
Multiple international bodies are working on AI governance norms, with varying levels of authority, participation, and effectiveness:
OECD AI Principles: The most widely adopted international AI principles framework. The OECD's 2019 AI Principles (Inclusive growth, Sustainable development, Human-centered values, Transparency, Security, Accountability, International cooperation) have been endorsed by G20 countries and dozens of others. They are non-binding but influential — shaping national frameworks and creating common vocabulary.
UN Governance Efforts: The UN General Assembly passed a non-binding AI resolution in 2024. The ITU (International Telecommunication Union) hosts the AI for Good platform. UNESCO's 2021 Recommendation on the Ethics of AI is the broadest-endorsed international AI ethics document. None of these create binding obligations.
G7 Hiroshima AI Process: G7 nations in 2023 agreed to a set of AI guiding principles and an International Code of Conduct for AI — voluntary commitments focused on trustworthy and safe AI development, with particular attention to frontier models.
Creating binding international AI governance faces fundamental obstacles. Countries disagree on what AI governance is for. The US prioritizes innovation and national security; the EU prioritizes fundamental rights; China prioritizes social stability; developing countries prioritize access and capacity building. These are not easily reconciled into binding treaty commitments.
AI capabilities evolve faster than treaty negotiation processes. The most powerful AI systems are developed by private companies that are not parties to international agreements. And unlike nuclear weapons or chemicals — where treaties have worked — AI systems have civilian dual-use potential that makes arms-control analogies imperfect.
Design a minimal viable binding international AI governance framework. Specify: who has authority, what is governed (specific AI capabilities or applications), how it is enforced, and how disagreements among member states are resolved.
Then identify the three biggest weaknesses of your design — what would most likely cause it to fail?
Every technical debate about AI governance sits atop deeper questions: who should have power in an AI-shaped world, what is AI for, and who gets to decide.
These questions don't have technical answers. They are political choices — and the countries, companies, and institutions making them right now are writing the first draft of the answer.
Behind the technical debates about AI regulation are deeper disagreements about what kind of AI future is desirable, who gets to decide, and how power should be distributed in an AI-shaped world. Three broad visions compete for influence in international AI governance discussions.
AI should be governed by rules that protect individual rights, maintain democratic accountability, preserve competitive markets, and prevent dangerous concentrations of power — whether in corporations or states. This vision underlies EU AI Act design and much of the OECD framework. It assumes that the primary risks of AI are to individuals and democratic institutions, and that governance should be structured to protect both.
AI is a strategic technology that states should guide for national development goals — economic growth, state capacity, national security, and social stability. AI governance should enable state use of AI while preventing its weaponization against the state or its deployment against social order. This vision underlies Chinese AI governance and resonates with some developing country governments that see state-led technology development as the path to economic catch-up.
AI's benefits should be shared globally, with developing countries gaining access to AI capabilities and a genuine voice in governance. AI governance should prevent both Western corporate dominance and Chinese state dominance, instead building genuinely multilateral institutions with real developing-country participation. This vision is advocated by many scholars and civil society organizations but has limited institutional power behind it.
In practice, AI governance norms are being set primarily by the EU (through law) and the US and China (through the behavior of their companies and the technology they export). The global commons vision exists primarily in academic papers and NGO advocacy. Understanding which vision is actually shaping AI governance requires watching what is being built, not just what is being said.
Choose a real AI governance document: the EU AI Act, NIST AI RMF, China's Generative AI Regulations, the G7 Hiroshima Code of Conduct, or OECD AI Principles.
Identify which of the three visions (liberal democratic, state-led development, global commons) most closely underlies its design. What specific provisions reflect that vision? What does it not address that another vision would prioritize?