China's AI Governance: From the AIDP to Companion AI Rules

China has established itself as one of the world's largest artificial intelligence economies and, at the same time, as one of the legal systems that has moved earliest to regulate specific applications of the technology. The report AI Governance around the World: Country Profile — China (July 2026), prepared by Arcangelo Leone de Castris, Rhoda Jiang and Miaowei Wang for the Alan Turing Institute, with expert review by Jason Zhu (Concordia AI), offers the most up-to-date portrait currently available of this regulatory architecture¹. This analysis reconstructs, corrects and expands on the report's content, situating it within the broader context of comparative AI law.

Unlike the European Union — which opted for a horizontal regulation, Regulation (EU) 2024/1689 (the AI Act) — China has built its framework through successive, sector-specific administrative rules aimed at applications considered high-risk: recommendation algorithms, deep synthesis (i.e., generative systems capable of producing synthetic audiovisual content), generative AI services and, since July 2026, anthropomorphic interaction services. The rest of the ecosystem — lower-risk applications — remains subject to considerably lighter oversight, delegated largely to the private sector and local administrations.

Chronology of the regulatory and strategic framework

Strategic objectives and high-level principles

The starting point of China's regulatory architecture is the **New Generation Artificial Intelligence Development Plan (AIDP, 2017)**², which positions artificial intelligence as the engine of the country's next cycle of economic development. The AIDP set out a roadmap with three milestones: reaching, by 2020, technological levels comparable to the world's most advanced economies; achieving, by 2025, substantive advances in basic AI theory that would turn the technology into a driving force of national industry; and establishing China, by 2030, as the world's leading center of AI innovation. To reach these goals, the AIDP articulates six cross-cutting objectives: building an open and collaborative innovation system, fostering a high-performing intelligent economy, achieving civil-military integration, developing ubiquitous intelligent infrastructure, planning major science and technology projects in advance, and building a serious and effective intelligent society. The Plan itself also recognized the need for a governance framework to accompany investment: the formulation of laws, regulations and ethical norms; the establishment of technical standards and intellectual property protection mechanisms; the creation of safety evaluation and oversight systems; and AI training and literacy programs.

On this basis, the National New Generation AI Governance Expert Committee³ — established under the Ministry of Science and Technology (MOST) and made up of representatives from industry and academia — published the 2019 Governance Principles for a New Generation of AI, which emphasize increasing the common wellbeing of humanity alongside safeguarding social security and respecting human rights. The principles set out — harmony and friendliness, fairness, inclusiveness, privacy, security, shared responsibility, open cooperation and agile governance — were developed further in 2021 through the Ethical Norms for New Generation Artificial Intelligence, which extend these principles across the full development, provision and use lifecycle of AI systems.

In October 2023, the Ministry of Foreign Affairs unveiled the Global AI Governance Initiative⁴, a document the original report omitted and which is essential to understanding the external projection of China's strategy. The Initiative calls for international cooperation on AI safety, respect for national sovereignty, opposition to technology monopolies and AI misuse, and greater representation of developing countries in multilateral policymaking processes; it even contemplates the possibility of centralized coordination under the United Nations umbrella.

Finally, in March 2024 China launched the "AI Plus" initiative, given concrete form through an August 2025 policy directive, setting a ten-year roadmap to integrate AI across the whole of the Chinese economy and society by 2035. Its intermediate milestones are ambitious: a 70% penetration rate of intelligent terminals and autonomous agents in key industrial and public sectors by 2027, and the deployment of AI applications across 90% of the general economy by 2030.

Taken together, these instruments reveal a three-phase evolution: an initial stage (2017-2019) centered on R&D investment and industrial capacity; a second stage (2019-2023) of ethical-governance soft law; and a third stage (2024 onward) that combines massive economic deployment with increasingly precise sectoral regulation. The thread running through all three phases is the recurring formula "development and security" (发展与安全), which expresses the Chinese state's determination to act simultaneously as promoter and regulator of the technology.

Definitions of the regulated technologies

China's legislator has so far declined to establish a general, cross-cutting definition of artificial intelligence, opting instead for ad hoc definitions within each sectoral instrument:

• Algorithmic recommendation technology (2022 Provisions): the use of generation, personalization, ranking, filtering or decision-making algorithms to supply information to users.
• Deep synthesis technology (2023 Provisions): technologies that use generative algorithms — with deep learning and virtual reality as leading examples — to produce text, images, audio, video or virtual environments.
• Generative AI technologies (2023 Interim Measures): models and related technologies capable of generating text, images, audio, video or other content.
• Machine learning algorithm (State Administration for Market Regulation technical standard, 2021): an ordered, finite set of rules that, through machine learning techniques, generates classifications, inferences or predictions from input data.

It is also worth noting, for its doctrinal significance, that the Standardization Administration of China (SAC) proposed in 2018 — in its White Paper on AI Standardization — a general, non-binding definition: AI as the set of theories, technologies, methods and application systems oriented toward using digital computers, or machines controlled by them, to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and apply it to achieve the best possible outcomes. This plurality of sectoral definitions, in contrast with the unitary model of Article 3(1) of the EU AI Act, directly shapes the scope of application of each Chinese regulatory instrument and explains why the coexistence of norms does not always clearly resolve cases of overlap between services.

High-risk application regime

China has favored specific regulation over horizontal legislation. In August 2023, the State Council announced its intention to draft a comprehensive AI Law⁵, still in drafting as of the time of writing (July 2026), together with rules on government data sharing and security. In the meantime, the Cyberspace Administration of China (CAC) has progressively built out four successive regimes:

Provisions on Algorithmic Recommendations (in force: March 1, 2022). Require providers to register their algorithms in a public registry maintained by the CAC, detailing training datasets and the security measures adopted. Services capable of "shaping public opinion" or "mobilizing society" must undergo risk assessments and periodic audits, whose documentation is added to the registry. The use of algorithms to manipulate public opinion, engage in anticompetitive practices, exploit user vulnerabilities or endanger national security is expressly prohibited; news service providers additionally require a specific license and must guarantee the accuracy of generated information.

Provisions on Deep Synthesis Services (in force: January 2023). Structured around four pillars — data security, transparency, content management and labeling, and technical security — these aim to prevent harm caused by deepfakes, requiring that users be able to identify when they are interacting with synthetic content, identity verification, and a feature database enabling detection of unlawful or harmful content.

Interim Measures for Generative AI Services (in force: August 2023). Set obligations on training datasets, labeling of generated content, personal data protection and user rights, including a complaints mechanism for accessing, downloading or deleting personal data. Providers must remove "unlawful content" and report it to the competent authorities; services capable of influencing public opinion are additionally subject to the safety assessments and algorithm registration required under the 2022 Provisions.

Interim Measures for Anthropomorphic AI Interaction Services (in force: July 15, 2026). This is the most recent rule and, in its subject matter, the most distinctive in the comparative landscape: it regulates systems that simulate personality, communicative style and sustained emotional interaction — artificial companionship, emotional support. The rule prohibits designing systems aimed at replacing social relationships, exercising psychological control or inducing emotional dependency, and requires providers to prevent their systems from generating content that encourages self-harm or suicide, damages users' real interpersonal relationships, or induces unreasonable decisions through emotional manipulation. Additional obligations cover personal data protection, risk warnings built into service design, and guidance on mental health and emotional boundaries. As far as the available comparative law allows one to determine, this is the first state regulation specifically directed at the relational and psychological risks of companion conversational AI, and it anticipates a regulatory terrain — so-called AI companions — in which other jurisdictions, including the European Union through the AI Act and child-protection legislation, have not yet developed an equally specific regime.

Local governance

Provincial and municipal administrations have developed their own frameworks, not always uniform in legal nature: some — Shanghai and Shenzhen — take the form of local regulations with normative effect; others — Beijing, Hangzhou and Guangzhou — are policy guidelines, action plans or implementation plans without equivalent binding force.

The Shanghai Regulation for Promoting the Development of the AI Industry (October 2022) introduces anticipatory regulatory mechanisms — AI sandboxes, public governance of computing capacity, co-design of standards with industry, and development of local standards — and requires high-risk systems to comply with the principles of necessity, adequacy and controllability, while leaving developers greater flexibility for lower-risk applications. The Shenzhen Regulation (November 2022), for its part, combines public financial support with a flexible risk-management framework: products classified as low-risk by the municipal administration can obtain certification even absent national or local standards, provided they meet equivalent international specifications; the regulation also requires the establishment of a municipal AI Ethics Committee.

Standardization system

China's technical standardization system is hierarchical and centrally coordinated by the state, organized across five levels — national, sectoral, local, industry-association and organizational — in which lower-level standards must be compatible with higher-level ones and may only be more stringent, never less. The Standardization Administration of China (SAC), part of the State Administration for Market Regulation (SAMR), coordinates national standardization plans, approves and publishes cross-cutting national standards, oversees subnational standardization activity, and represents China in international standard-setting bodies.

Standards may be mandatory — carrying the force of law when they protect health, safety or intellectual property — or recommended; however, voluntary standards are frequently cited or referenced by regulation, thereby acquiring the status of a de facto requirement. Technical committees include government, industry and academic representation — including foreign entities — although civil society involvement remains poorly documented.

AI-specific standardization is driven, at the operational level, by the China Electronics Standardization Institute (CESI), attached to the Ministry of Industry and Information Technology (MIIT) and tasked with coordinating the objectives set by the National AI Standardization Group. CESI has no authority to approve standards, but it organizes the technical committees and working groups whose output is subsequently adopted by MIIT as sectoral standards. As of June 2026, more than fifty national AI standards had been published, with a similar number in development⁶.

China's national AI standardization agenda was formalized in July 2020 through the Guidelines for the Construction of a National New Generation Artificial Intelligence Standard System, updated in June 2024 following a public consultation process. The 2024 version sets, for 2026, the goal of surpassing fifty national and sectoral standards, achieving adoption by more than a thousand companies, and participating in at least twenty international standardization projects, reorganizing the system into seven domains: common foundations, foundational support, key technologies, intelligent products and services, capacity-building for new industrialization, sectoral applications, and safety/governance. This last domain, classified in 2020 as a cross-cutting "safety/ethics" category, became in 2024 an autonomous structural domain — evidence of an increasingly consolidated understanding of technical standardization as a governance instrument rather than a mere adjunct to product quality.

Among the most recent regulatory milestones are the February 2024 publication, by the SAC's Cybersecurity Technical Committee (SAC/TC260), of the Basic Safety Requirements for Generative Artificial Intelligence Services — China's first standard on generative AI, which builds on the 2023 Interim Measures and includes an annex listing more than thirty risks characteristic of these systems; the September 2024 publication of the AI Safety Governance Framework, a non-binding guidance document; and the December 2024 creation, by MIIT, of a dedicated AI standardization Technical Committee, replacing the previous division of responsibilities among Committees TC28 (information technology), TC260 (cybersecurity) and TC124 (electronic measuring instrumentation). This institutional reorganization reflects AI's transition from a subcategory of ICT to a standalone, cross-cutting regulatory domain.

Internationally, China has moved from a defensive posture — focused on developing domestic standards — toward an active and growing role in international standard-setting bodies, with official quantitative indicators tracking chairmanships, secretariats and registered experts within those bodies, the number of international standards adopted domestically, and international meetings held on Chinese soil. The government explicitly conceives of this projection as a lever of geopolitical influence and international economic competitiveness.

Critical assessment

The Chinese model combines regulatory agility with structural limitations worth bearing in mind. Sectoral administrative rules — provisions and interim measures — allow for a faster, more technically specific response than drafting a law, but within China's normative hierarchy they rank below laws and administrative regulations proper. As a result, they lack the authority to curtail citizens' or legal persons' rights, increase their obligations, expand departmental powers, or reduce pre-existing legal duties without support from a higher-ranking norm. This architecture explains both the speed of China's regulatory response to phenomena such as deep synthesis or anthropomorphic interaction, and the persistence, nearly a decade after the AIDP, of a comprehensive AI Law still in drafting.

From a comparative law perspective, the contrast with the European Union is instructive: while the AI Act starts from a risk classification applicable horizontally to any AI system, the Chinese legal order builds vertical regimes, defined application by application, without a general category equivalent to the "high-risk systems" of Article 6 of the AI Act. This approach delivers sectoral technical precision, but also generates boundary frictions between rules — for example, between the 2022 Provisions and the 2023 Measures when a recommendation service incorporates generative components — that China's legislator has not resolved systematically.

Conclusion

China's regulatory trajectory, from the 2017 AIDP to the 2026 Interim Measures for Anthropomorphic Interaction Services, reveals a deliberately incremental strategy: administrative regulation moves ahead by application area, while the higher legislative framework — the long-announced comprehensive AI Law — remains pending. Added to this is a strongly state-coordinated technical standardization system, increasingly focused on its international projection as an instrument of competitiveness. For legal practitioners analyzing the phenomenon from jurisdictions with a different regulatory tradition — such as the European or Spanish ones — the Chinese case offers a valuable point of comparison: it demonstrates that it is possible to regulate specific-risk applications quickly and with granularity without a prior horizontal law, at the cost of a fragmented normative architecture whose internal consistency will ultimately depend on whether the promised AI Law finally comes to fruition.

Notes and references

1. Leone de Castris, A., R. Jiang & M. Wang (2026), AI Governance around the World: Country Profile — China, The Alan Turing Institute, available at: https://doi.org/10.5281/zenodo.21031557.
2. State Council of the People's Republic of China, New Generation Artificial Intelligence Development Plan (新一代人工智能发展规划), 2017.
3. National New Generation Artificial Intelligence Governance Expert Committee, Governance Principles for a New Generation of AI, 2019; Ethical Norms for New Generation Artificial Intelligence, 2021.
4. Ministry of Foreign Affairs of the People's Republic of China, Global AI Governance Initiative (全球人工智能治理倡议), 2023.
5. State Council of the People's Republic of China, Legislative Work Plan (国务院办公厅关于印发国务院2023年度立法工作计划的通知), 2023.
6. National Public Service Platform for Standards Information (China), accessed June 21, 2026, as cited in the Alan Turing Institute report.