I would say: Comrades, welcome to Shanghai! I believe that solidarity should be the basis for us to successfully achieve digital sovereignty. Because I myself am a legal expert and law professor, my speech will focus on law but I will use ordered innovation to describe China's approach to AI regulation.
Throughout the world, there is a typical saying that there are three models of AI-related law. One book Digital Empires – The Global Battle to Regulate Technology, written by Anu Bradford, law professor at Columbia Law School, shows the three types of models. First, the United States mainly pioneers a 'market-driven model.' China is considered a 'state-driven model' and EU a 'rights-driven model'. When we are talking about digital sovereignty, we need to consider the role of the state in the whole scenario. For me, this typology is based on ideology rather than detailed analysis of the legal system in different countries. For me, EU is not a rights-oriented system. Actually, AI Act anchors to TFEU Art.114, which considered market harmonisation, namely the single market within EU as its major legislative objective. EU actually shows a risk-aversion model. The United States shouldn't be considered as a market-based model because there is no law on the federal level in the United States to protect, for example, personal information. All the laws passed by the US Congress related to AI have been targeting its competitors. But now, the competitor from the perspective of United States would always be China. We know the TikTok Law used national security as the general justification for its legislation. The United States may not have laws to protect its citizens in the digital world or to deal with specific problems arising from the digital period. All the laws in the United States on the federal level related to digital technology have been targeting foreign countries on the basis of the so-called national security.
Table 1: Global Artificial Intelligence-Related Legislation Models
| Mode | Core Value Orientation | Legislative Goals | Primary Legislative Tools & Features | Representative Practices & Cases |
|---|---|---|---|---|
| China Model | Ordered Innovation | Balancing technological development and risk control, enhancing industrial competitiveness | Progressive legislation, local experimental rules (e.g., fault-tolerant mechanisms), phased planning; emphasising public-private collaboration and institutional flexibility | Next-Generation Artificial Intelligence Development Plan (Three-Step Strategy), local regulations (e.g., Shenzhen's AI Industry Promotion Ordinance) |
| US Model | Pan-Securitisation | Maintaining technological hegemony, addressing geopolitical challenges | Domestic light-touch regulation (relying on industry self-regulation and soft law), external hardline sanctions (export controls and investment screening); security exception mechanisms | CHIPS and Science Act (technology blockade), Foreign Investment Risk Review Modernisation Act (foreign investment screening) |
| EU Model | Risk Aversion | Minimising social risks, safeguarding fundamental rights | Pre-emptive preventive rules (high-risk classification), mandatory compliance (e.g., impact assessments), ethics-driven legislation; emphasising regulatory redundancy | GDPR (Data Protection Impact Assessment), AI Act (four-tier risk classification and prohibited list) |
The concept of national security in the United States is different from that in all the other countries, where national security is used to defend their own sovereignty. But in the United States, national security is to maintain its supremacy or dominance in the world. Therefore, the concept of national security of the United States is damaging world peace. Therefore, I would say the US model is a pan-securitisation one, namely, using national security to do anything.
II. Core Value Orientations
| Model | Driving Logic | Key Mechanism |
|---|---|---|
| EU | Risk Aversion | • 'Dangerous product' classification • Ex-post fines over innovation incentives |
| US | National Security Hegemony | • H.R.1's state pre-emption clauses • Algorithmic warfare export controls |
| China | Ordered Innovation | • Staged legislation (2020/2025/2030) • Fault-tolerant local rules • Ethics-by-design (Art. 14 GenAI Measures) |
The Chinese model shouldn't be considered as 'state-dominant.' We found out that although China has yet to make a law called 'AI Law,' it actually started to draft the AI law 2 years ago and has postponed the making of that law. However, if we look at Chinese laws, there exists a comprehensive system. First of all, we have three laws targeting digital technology. One is the Cybersecurity Law enacted in 2017. The other two laws were both enacted in 2021; they are the Personal Information Protection Law and Data Security Law. And below these laws, there are a number of regulations dealing with every aspect of digital technology and AI specifically. For example, China issued regulations on recommendation algorithms, deep synthetic algorithms, and generative AI. Therefore, in China, we found a comprehensive system of laws related to AI in order to protect different legal interests, from personal rights, to national security, and to innovation.
| Category | Core Content | Representative Cases/Examples | Key Features |
|---|---|---|---|
| Technology-Specific Regulation | Classified supervision of AI technologies/applications, emphasising entity accountability and dynamic adjustment. | Internet Information Service Algorithmic Recommendation Management Provisions (2021), Deep Synthesis Management Provisions (2022), Generative AI Service Management Measures (2023) | Agile governance: Rules dynamically updated to match technological evolution; prohibits algorithmic abuse (e.g., price discrimination, addiction induction). |
| Local Industrial Promotion Laws | Stimulates innovation and economic empowerment through institutional support. | Shanghai's AI Industry Development Promotion Regulations (2022): Pioneered 'error-tolerance mechanisms'; Shenzhen's AI Industry Promotion Regulations (2023): Established dedicated funding and ethics committees | Deregulation focus: Infrastructure co-development (e.g., smart computing clusters), public data sharing, and reduced pre-approval burdens. |
| Rights & Security Safeguards | Protects national/personal rights by regulating computing carriers (networks), algorithmic fuel (data), and subjects (personal information). | Cybersecurity Law (2016), Data Security Law (2021), Personal Information Protection Law (2021) | Closed-loop constraints: Covers all technical elements (computing power, algorithms, data); grants individuals algorithmic opt-out and explanation rights. |
| Ethical Governance System | Translates ethical principles (e.g., 'human-centred AI') into enforceable compliance duties embedded in development lifecycles. | Art. 4 of Generative AI Service Management Measures (requiring adherence to social ethics); Shenzhen Regulations Art. 21 (mandating corporate ethics committees) | Context-adaptive implementation: Prioritises preventive ethical constraints with 'human-in-the-loop' mechanisms (e.g., manual intervention), differing from EU's rigid checklists. |
| Industry Standards Framework | Bridges technical specifications and legal norms, promoting innovation and global collaboration. | National AI Industry Standardisation System Guidelines, Shanghai Regulations Art. 66 (empowering ethics committees to draft guidelines) | Ecological shift: Integrates carbon footprint tracking and energy efficiency metrics, forming a 'standards-laws-policies' tripartite model. |
| Technical Elements Coverage | Governs AI's three pillars: computing power, algorithms, and data. | Computing: Critical Information Infrastructure Security Protection Regulations (2021), 'East Data West Computing' project; Algorithms: Ministry-level rules; Data: Cross-Border Data Security Assessment Measures (2022), Zhejiang's public data regulations | Comprehensive oversight: Optimises computing layout via policy incentives; constructs compliant data circulation paths. |
| Japanese Reference Case | Japan's draft AI Promotion Act mirrors China's 'ordered innovation' logic. | Emphasises national strategic planning, multi-stakeholder collaboration, and balanced risk-control innovation. | Cross-jurisdictional validation: Confirms 'ordered innovation' as a replicable East Asian governance paradigm prioritising developmental pragmatism over ideological rigidity. |
This is my typology for this issue. China represents an ordered-innovation model with different laws to address different concerns in the society. The US model for me is a pan-securitisation model, which is trying to maintain technological hegemony and address geopolitical challenges. The EU model is a risk-aversion model which transplants its legal ideas in the product security field into AI regulation, as exemplified by its risk level methodology.
We found out that if the United States' model became predominant, there would be nothing left for Global South countries because it would not allow any challenges to its dominance in technology, economy, and so on.
The Chinese model actually opens a space for the Global South. Just a couple of days ago, we discussed China's open-source generative AI such as DeepSeek and a number of other open-source models. According to a statistical report provided by Hugging Face, a widely used platform in the AI field, among the world's top ten open-source LLM models, nine are from China. Open-source AI models are just one of the many ways to show how the system would be open to the whole world. For example, Global South countries such as Brazil, Congo or Malaysia, can develop their own AI applications on the basis of the foundational AI models which are open source and created by Chinese innovators. With this open-source movement, we can foresee that the world in the digital era can move towards a more equitable, dialogic and progressive future. Actually there are many legal details that I do not have enough time to discuss. I just outline my general theoretical framework for AI governance.
I would like to talk a little bit about the conclusions to finish my speech. First, China's AI governance model is a viable alternative for Global South countries, challenging the dominant western neoliberal development framework, with the United States dominating in the field as the foundation of the old system. Secondly, China's institutional elasticity characterised by phased legislation, local experimentalism and adaptive sequencing, effectively balances technological innovation with risk control, avoiding the regulatory rigidity of the EU and the security-driven containment of the United States. And finally, this approach offers the Global South a pragmatic, context-sensitive development pathway that prioritises sovereignty, capability building, and technological self-determination over an imported universal blueprint.
Watch the whole speech here: