This is the second part of an analysis of China and global cyberspace governance.
THE ROOT cause of the current global cyberspace governance predicament of artificial intelligence (AI) lies in the fact that existing governance paradigms remain anchored in the era of the “information interaction space,” creating an ontological mismatch with the generative, evolutionary and ecological logic of the intelligent action domain.
Clinging to static rules, territorial jurisdiction and centralised authority will inevitably result in systemic governance failure.
In this context, there is an urgent need to construct a governance paradigm that can effectively respond to the mechanisms of action generation, the rhythm of system evolution, and the characteristics of the power ecology, namely “generative governance.”
Rather than merely adding regulatory tools to existing frameworks, generative governance seeks to realign with the ontological transformation of cyberspace by reconstructing governance objects, pathways and conditions.
Its core orientation includes three shifts: first, shifting the focus of governance from behavioral subjects and isolated events to the mechanisms of action generation; second, shifting governance approaches from ex post correction to ex ante shaping; and third, shifting governance conditions from single-authority dominance to multi-node ecological coordination.
Taking shifts
The ultimate goal of generative governance is not to remedy abnormal behaviour after it occurs but to shape model structures, system interfaces and ecological boundaries in ways that encourage the intelligent action domain to produce behaviors that are governable, auditable and tunable -
Philosophical shift: The philosophical foundation of generative governance lies in conceiving governance not as an external force imposing limits on existing behavior but as a process of creating internal conditions within systems to guide action generation toward governable directions. In the intelligent action domain, actions exhibit strong generativity, emerging from the interaction of data, model architectures, feedback loops and algorithmic parameters. If governance continues to treat actions as direct expressions of subjective intent and formulates rules accordingly, it will inevitably face challenges such as ambiguous agency, untraceable intent and the distributed nature of action chains.
Generative governance therefore requires a shift from a behaviour-attribution philosophy to a structural-generation philosophy. Under this framework, the central task of governance is no longer merely to define legal and illegal actions but to determine which generative mechanisms are permissible, which must be restricted, and which should incorporate external constraints. Questions such as whether models possess explainability, whether audit interfaces are embedded, and whether real-time monitoring is supported become concrete manifestations of governance philosophies rather than technical details. In essence, the foundation of governance shifts from ex post accountability to ex ante shaping.
At the same time, generative governance recognises the plurality of governance actors. Because the logic of action generation is distributed across models, platforms, computing infrastructures and technological ecosystems, governance cannot be accomplished by any single actor alone in the intelligent action domain. States, platform enterprises, model developers, computing power providers and open-source communities all participate in shaping the action domain and must therefore collectively contribute to governance.
Reconstruction of governance logic: Reconstructing governance logic requires clarifying how governance should intervene in the process of action generation rather than simply determining who governs.
In traditional governance frameworks, the logic of governance follows a linear structure, namely “rule-making-behavior identification-enforcement intervention-ex post accountability.” This model presupposes identifiable behaviors, relatively stable action chains and clearly separable consequences.
Within the intelligent action domain, however, behaviours exhibit generative characteristics. An action chain may originate within a model, be amplified on a platform, and produce consequences within another ecosystem. The entire process is difficult to pause, reproduce, or neatly segment. If governance continues to focus only on the end stage of behavior, it will remain trapped in reactive remediation. Generative governance therefore shifts attention from behaviour itself to how behaviour is generated, moving from single-point regulation to the co-shaping of generative conditions. At its core lies the guidance or inhibition of action generation by shaping the preconditions and evolutionary paths through which such actions emerge.
Governance must also intervene earlier in the action chain, for example, in the selection of training data, parameter adjustments, and alignment strategies during model development; in information flow logic, content labeling mechanisms, and algorithmic transparency during platform operation; and in interoperability frameworks and open-source governance standards at the ecosystem level.
These factors not only form part of the logic behind action generation but also serve as critical points of intervention for governance.
An ongoing process
Moreover, this governance logic must remain dynamic and iterative. As generative mechanisms evolve, governance must continuously adjust through evaluation, feedback, and redesign. Governance thus becomes not a one-time issuance of rules but an ongoing process of calibrating generative conditions so that governance rhythms remain aligned with the evolutionary pace of the intelligent action domain.
Reconstructing governance structure: The governance structure concerns governance subjects, power distribution and institutional design. The philosophical and logical shifts of generative governance ultimately require institutional implementation through structural transformation.
Although traditional global cyberspace governance involves multiple stakeholders, its underlying power structure still centers on state authority. International organisations produce normative principles. National governments translate them into domestic regulations and enforcement. Platforms and technical communities operate within these legal frameworks. The overall structure therefore demonstrates a top-down hierarchy and sovereignty-centered characteristics.
In the intelligent action domain, however, key nodes determining how actions are generated, such as model architectures, algorithmic logic and ecosystem interfaces, are increasingly controlled by transnational technology companies, model R&D institutions, open-source communities and infrastructure operators. These nodes exercise generative and system-embedded power that often precedes or operates independently of state legislative and enforcement processes.
Consequently, governance structures must shift from hierarchical transmission models toward multi-node, ecosystem-based generative coordination. In this new framework, multiple actors jointly participate in governance; states define fundamental values and legitimacy boundaries; platforms establish trustworthy information flow environments; model developers embed auditability and value alignment into system architectures; computing power providers ensure infrastructure security; and technical communities promote open, transparent and verifiable innovation mechanisms. Each actor functions both as a governance participant and as an object of oversight by other actors.
At the same time, the boundaries of governance also transform. Governance units are no longer defined primarily by territorial borders but increasingly by technical boundaries, such as technology stacks, model families, protocol standards and platform architectures. As long as governance logic within a technological ecosystem remains coherent, coordination can occur even when participants belong to different sovereign jurisdictions. In this sense, the reconstruction of governance structures implies a shift in the reference frame of governance from territorial sovereignty to technological ecosystems. — Contemporary World Magazine
Cai Cuihong is a professor at Centre for American Studies, Fudan University. Contemporary World Magazine is an international relations and social science journal published in Beijing, China.
