China Global AI Governance Initiative

Overview

The Global AI Governance Initiative is a policy proposal issued by the Government of the People’s Republic of China for international cooperation on artificial intelligence. The Ministry of Foreign Affairs publishes its full text with an update date of October 20, 2023. The Initiative presents AI as a source of substantial opportunities for socioeconomic development that also introduces unpredictable risks and complex challenges. It calls for equal attention to development and security, international dialogue, and open, fair, and efficient governance mechanisms.

The Initiative is non-binding policy, not legislation, a binding regulation, or an international treaty. Its adopted status denotes an issued policy rather than legislative enactment or an instrument awaiting signature. Its language principally expresses recommendations, calls for cooperation, and positions on responsible AI development. It establishes no regulator, licensing system, enforceable compliance deadlines, penalties, or appeal procedure. Organizations can use its principles for voluntary policy alignment while distinguishing those principles from obligations under separately applicable laws. The document specifies no legally operative commencement date, implementation deadline, or expiration date.

Scope and Governance Concepts

The Initiative has a global policy orientation but does not create a directly applicable worldwide legal regime. Its intended participants include governments, international organizations, companies, research institutes, civil organizations, and individuals. Its subject matter encompasses AI research and development, products and services, deployment and application, testing and assessment, ethics and governance, international technological cooperation, and capacity building. It does not establish statutory jurisdiction, covered-entity thresholds, or a technical or legal definition of artificial intelligence.

The document uses broad policy terminology rather than statutory coverage tests. It does not define separate categories for general-purpose models, generative systems, foundation models, autonomous systems, or particular deployment sectors, and sets no thresholds based on computing resources, user numbers, market presence, or organizational size. Risk-level testing, agile governance, and tiered and category-based management describe proposed governance approaches without establishing formal classifications. Explainability, predictability, human control, reviewability, monitoring, and traceability are desired characteristics rather than certification criteria, minimum performance requirements, or terms governed by a legally authoritative glossary.

Governance and Institutional Framework

The governance model emphasizes broad participation and consensus under principles of extensive consultation, joint contribution, and shared benefits. It encourages policy communication, information exchange, technological cooperation, and respect for differences among national policies and practices. Governments, international organizations, businesses, research institutes, civil organizations, and individuals are invited to participate. This cooperative approach does not allocate regulatory powers to a designated Chinese agency, and publication by the Ministry of Foreign Affairs does not confer enforcement jurisdiction over AI developers or deployers.

At the international level, the Initiative supports discussions within the United Nations framework about establishing an institution to govern AI and coordinate major issues concerning development, security, and governance. This proposal does not itself establish an organization, treaty, membership system, or budget. The Initiative also advocates ethical principles, accountability mechanisms, ethical guidelines, and science and technology ethical review and regulatory arrangements. It encourages clarification of responsibilities and power boundaries among AI-related entities and timely attention to ethical concerns, without specifying an institutional mandate, appointment process, reporting hierarchy, or compulsory review procedure.

People-Centered Development and Responsible Use

The Initiative advocates a people-centered approach that improves human wellbeing, respects human rights and interests, and maintains social security. It connects AI development with shared values and beneficial applications rather than technological advancement alone. Sustainable development is an explicit objective, including applications addressing climate change and biodiversity conservation. Development and security are presented as complementary priorities: AI should produce social and economic benefits while governance addresses the risks associated with its creation and use.

National sovereignty is another central principle. The Initiative calls for respect for other countries’ sovereignty and applicable national laws when supplying AI products and services abroad, as well as respect for relevant international law. It opposes using AI to manipulate public opinion, spread disinformation, interfere in other countries’ affairs, or jeopardize sovereignty. It urges cooperation against misuse by terrorists, extreme forces, and transnational organized criminal groups. Military AI should be approached prudently and responsibly. These positions express policy expectations without establishing a military treaty, export-control arrangement, or international enforcement mechanism.

Risk Management and Trustworthy AI

The Initiative proposes testing and assessment based on AI risk levels, agile governance, and tiered and category-based management. These approaches are intended to support adaptable responses as technologies and risks evolve. The document does not prescribe mandatory risk classifications, assessment procedures, a testing schedule, or an implementation manual. Countries are encouraged to establish or improve relevant laws, regulations, rules, and governance arrangements while building international consensus through dialogue. The design and sequencing of subsequent measures remain matters for participating governments and institutions.

Research and development entities are encouraged to improve explainability and predictability, enhance data authenticity and accuracy, and keep AI under human control. Trustworthy technologies should be capable of review, monitoring, and tracing. The Initiative also supports developing and applying technologies for AI governance, including using AI to prevent AI risks. These recommendations do not prescribe a particular oversight architecture, technical metric, certification standard, record format, registration filing, named compliance officer, or mandatory third-party audit. Voluntary alignment can include proportionate risk assessment, human oversight, data-quality review, and documentation of organizational decisions.

Privacy, Fairness, and Ethical Accountability

The Initiative calls for privacy and data security safeguards and opposes illegal collection and use of personal information. It places fairness and non-discrimination across the AI lifecycle, including data collection, algorithm design, technology development, and product application. Data authenticity and accuracy are linked to trustworthy development and use. These principles support examination of how data practices and system design may affect individuals and groups, while the Initiative itself does not establish a separate personal-information compliance regime or specify technical privacy controls.

Ethical accountability includes developing ethical guidelines, clarifying responsibilities, addressing ethical concerns promptly, and creating science and technology ethical review arrangements. Organizations pursuing voluntary alignment can evaluate discrimination risks, examine privacy and security safeguards, and identify responsibility for governance decisions. Such measures are practical applications of the policy rather than Initiative-specific legal duties. The call for accountability does not create civil liability standards, compensation rights, insurance requirements, a burden of proof, or a self-contained liability system for developers, deployers, governments, or other participants.

Monitoring, Evaluation, and Legal Consequences

Risk-based testing and assessment, reviewability, monitoring, and traceability support an adaptive approach to evaluation rather than a single review at deployment. The Initiative encourages attention to technological advances, risk assessments, policy communication, and exchanges of best practices. It does not establish a formal supervisory program, assessment frequency, audit standard, incident-reporting deadline, regulator notification channel, or mandatory performance indicator. It also does not require independent assessments or publication of assessment results. Organizations may document their chosen methods and responsibilities as part of voluntary alignment.

The Initiative contains no fines, criminal offenses, administrative sanctions, licensing consequences, or other enforcement penalties. It creates no cause of action, complaint body, appeal route, judicial-review mechanism, or dispute-resolution procedure. Its opposition to harmful practices and unlawful personal-information handling does not create new offenses. Its express support for respecting national laws and relevant international law means that AI activities remain subject to independently applicable legal rules. Policy alignment therefore needs to be distinguished from legal compliance in the relevant jurisdiction and destination market; the absence of penalties in this policy is not an exemption from other laws.

Relationship to Other Instruments

The Initiative calls for laws, regulations, rules, ethical guidelines, and governance standards to be developed or improved. It does not amend a named statute, incorporate a specific national AI regulation by reference, establish a legal hierarchy or conflict rule, or exempt participants from existing requirements. Its references to national laws and relevant international law are general. It functions alongside independently applicable legal regimes, rather than replacing them, and does not enumerate a detailed set of Chinese data, cybersecurity, or AI compliance obligations.

The Global AI Governance Action Plan, published on July 26, 2025, is a related subsequent policy instrument. It addresses infrastructure, open-source ecosystems, data, environmental sustainability, standards, safety governance, and capacity building in greater operational detail. It also refers to the United Nations Pact for the Future and Global Digital Compact. Those references and additional evaluation and safety proposals belong to the Action Plan, not to the 2023 Initiative. The Action Plan does not expressly replace the Initiative, and the two documents should be considered separately when identifying their respective policy proposals.

International Cooperation and Inclusion

International cooperation is foundational to the Initiative. It promotes shared AI knowledge, technological exchange, open-source availability, and equitable access to AI development and use. It opposes ideological divisions, exclusive groupings, technological monopolies, and unilateral coercive measures that obstruct development or disrupt the global AI supply chain. These positions are policy preferences, not trade rules or legally enforceable market-access guarantees. Cooperation is framed around consultation, shared benefits, and respect for differences among countries, rather than a binding harmonized standard.

The Initiative places particular emphasis on increasing developing countries’ representation and voice, ensuring equal rights and opportunities, and providing assistance to reduce disparities in AI development and governance capacity. Capacity building and inclusive participation are linked to the wider objective of making AI benefits accessible across countries. Support for United Nations discussions supplies an institutional direction without creating a UN body or international obligation. The Initiative does not establish mutual recognition, certification equivalence, or cross-border supervisory arrangements. Its international alignment is cooperative and aspirational, with national sovereignty and consensus remaining central principles.

Implementation Timeline

Voluntary Alignment Checklist

Sources and References