On 17 July 2026, at the 2026 World AI Conference (“WAIC”) held in Shanghai, China’s National Development and Reform Commission (“NDRC”), together with the Ministry of Education, the Ministry of Science and Technology, the Ministry of Industry and Information Technology, the National Data Administration, China Media Group, and others, jointly released the Action Plan for Artificial Intelligence Cooperation and Development (“Action Plan”). China’s Action Plan may be viewed as a transnational rule-making proposal carried by “cooperation”. It does not create new hard international law; rather, through eight actions—including the establishment of cross-border trusted data spaces, responsible open-source security guidelines, a framework for regulated application of agents, and soft standards on algorithmic non-discrimination—it exports a set of “legal modules” capable of being embedded into bilateral agreements, technical standards, and contractual clauses.
World Artificial Intelligence Cooperation Organization (WAICO)
On 16 July 2026, one day prior to the release of the Action Plan, 29 countries signed the Agreement on the Establishment of the World Artificial Intelligence Cooperation Organization in Shanghai, formally establishing WAICO with its permanent headquarters in Shanghai. This is the world’s first intergovernmental international organization dedicated exclusively to artificial intelligence cooperation. Its founding member states include China, Russia, Kazakhstan, Pakistan, Indonesia, Brazil, Venezuela, Laos, 10 African countries, and 12 Asian countries, among others. UN Secretary-General António Guterres attended the signing ceremony.
The establishment of WAICO carries the following significant implications:
1. A Breakthrough in the Law of International Organizations: WAICO was established by an “agreement” rather than a “treaty”, thereby lowering the threshold for domestic ratification by member states and reflecting a pragmatic design that facilitates participation by developing countries. Its organizational charter—including provisions on legal personality, privileges and immunities, and voting mechanisms—will serve as a new subject of study in the law of international organizations.
2. Institutional Empowerment of the Action Plan: The Action Plan is a policy document unilaterally released by China at the WAIC. Its specific commitments—such as providing 5,000 AI training slots to developing countries over the next five years and deploying the “Mazu” meteorological early warning system in 30 countries—are closely aligned with WAICO’s functions and may be viewed as the “start-up funding” and “initial projects” contributed by China to WAICO. This elevates the Action Plan from a unilateral policy declaration to an implementation commitment linked to a multilateral organizational framework.
3. Institutionalization of Rule Competition: While the EU projects its rules through the AI Act [ https://eur-lex.europa.eu/eli/reg/2024/1689/oj ] and the United States exerts pressure through bilateral digital trade agreements and export controls, WAICO is the first dedicated international AI organization initiated by China and headquartered in a Chinese city. It marks a shift in China’s AI governance pathway from “proposing rules” to “institutionalized rule export.”
Notably, India has not joined WAICO. This may affect the organization’s representativeness in coordinating AI governance in the Asia-Pacific region and leaves a variable that requires separate handling in the compliance planning of Chinese enterprises in that market.
The Action Plan
The Action Plan is anchored in the principle that “artificial intelligence should be an international public good that benefits all humanity.” It addresses data, computing power, ecosystems, empowerment, talent, rules, governance, and ethics, setting forth eight actions: Quality Data Supply, Inclusive Intelligent Computing Power, Open-Source Ecosystem Sharing, Deep AI Empowerment, Joint Cultivation of Digital and AI Talents, Joint Development of Rules and Standards, Security Governance Collaboration, and AI for Good. These actions respond to United Nations initiatives to strengthen international AI cooperation, bridge the digital divide, and leverage AI for sustainable development. The most legally impactful actions are outlined below.
1. Quality Data Supply: This action proposes to “promote cross-border data flows, build and operate cross-border trusted data spaces in certain sectors, and facilitate efficient, convenient, and secure cross-border data flows. It also aims to collaboratively construct high-quality corpora and industry-grade datasets, promote multilingual corpus co-construction and sharing, and lay a solid foundation for global AI innovation.” The “cross-border trusted data spaces” represents a “third way” distinct from the EU’s adequacy decisions and standard contractual clauses, and the U.S. model of free data flows. Its legal significance lies in deploying technical controls and contractual trust on a “space” basis, potentially giving rise to a new type of “special regulatory sandbox” for cross-border data. More bluntly, the “whereas” clauses and compliance schedules of transnational data contracts will soon need to answer whether, and how, to connect to such spatial architectures advocated by China. This is highly likely to become a focal point of negotiation in the governing law and dispute resolution clauses of data cooperation agreements.
2. Open-Source Ecosystem Sharing: This action proposes to “encourage the joint development of an international AI open-source community, carry out international exchanges and cooperation among open-source communities, and promote the sharing of general-purpose large models, foundational algorithms, and tool components. It calls for the collaborative development of open-source compliance systems and security guidelines, supports localized innovation by various countries based on open-source models, and seeks to build an open, shared, secure, orderly, and collaboratively governed global AI open-source ecosystem.” The phrase “collaboratively develop open-source compliance systems and security guidelines” goes beyond the scope of traditional open-source licenses by juxtaposing “security” and “compliance”. This signals the emergence of a “responsible open-source” standard that transcends traditional knowledge-sharing rules such as the Apache and GPL licenses, meaning that in the open-source model supply chain, security review and compliance commitments may soon shift from being optional to becoming prerequisites for market access.
3. Deep AI Empowerment: This action proposes to “deepen ‘AI+’ cooperation, establish transnational industry cooperation platforms, support digital and AI capacity building in developing countries, promote the regulated application and innovative development of agents, and advance AI application and empowerment in fields such as science, manufacturing, healthcare, education, agriculture, and governance, so as to use AI to promote economic development, improve social governance, and enhance people’s well-being.” By expressly mentioning “promote the regulated application and innovative development of agents”, this is the first time “agents” are given independent expression in a national-level international cooperation document. Its significance is that agents are moving from being mere tools to objects of legal concern with autonomous decision-making capabilities. This essentially opens the first window for international coordination on the legal vacuum of “agent liability”. As agents enter fields such as manufacturing, healthcare, and governance, their “regulated application” may evolve into mandatory standards covering algorithm filing, behavioral boundaries, intervention rights, and termination mechanisms, directly impacting product liability and professional ethics rules.
4. Joint Development of Rules and Standards: This action proposes to “jointly build AI standards and specification systems, promote the development and revision of international standards, collaboratively construct standard alignment and coordination mechanisms, and foster inclusiveness and interoperability among standard systems. It also calls for strengthening the alignment and coordination of AI development strategies, governance rules, and technical standards.” This is the part of the Action Plan with the greatest legal leverage. It seeks to replicate the successful experience of the telecommunications sector by embedding Chinese-led technical and security standards into global AI supply chain compliance through a “standards-essential” pathway.
5. Security Governance Collaboration: This action proposes to “jointly build AI security governance mechanisms, strengthen cybersecurity threat information sharing and emergency response cooperation, and prevent the misuse and abuse of AI technologies. It promotes research to enhance AI explainability, transparency, and security, strengthens AI data governance, and fosters a sound environment for global AI development. It advocates for establishing open platforms to share best practices and promote international cooperation on AI security governance globally.” Once “enhancing AI explainability and transparency” is incorporated into international cooperation mechanisms, it will directly translate into evidentiary rule challenges in cross-border litigation. In cases involving algorithmic discrimination or harm from automated decision-making, plaintiffs may rely on this to assert that defendants bear an “explainability obligation,” with the standard of that obligation referencing Chinese-led international security guidelines. This would elevate “explainability” from a technical ethics term to a legal standard of due diligence, triggering disputes over the cross-border discovery and admissibility of novel forms of evidence such as algorithm audit reports and model cards.
6. AI for Good: This action proposes to “uphold the AI scientific and technological ethical principles of openness and transparency, privacy and security protection, and controllability and trustworthiness, and jointly build a system of AI ethical guidelines. It implements the people-centered and AI-for-good philosophy to promote the building of a warmer intelligent society. It aims to eliminate racial discrimination, other forms of discrimination, and algorithmic bias, and to safeguard fairness and non-discrimination. It advances international governance cooperation on AI scientific research, contributes public science products for the Global South, and serves the UN 2030 Sustainable Development Goals.” By proposing to “eliminate racial discrimination, other forms of discrimination, and algorithmic bias, and safeguard fairness and non-discrimination” while contributing public goods for the “Global South,” this action aligns itself with both the UN 2030 Agenda and international human rights law discourse. It may become a “soft mandatory standard” against which the compliance of AI products exported by Chinese enterprises is judged. NGOs and competing jurisdictions may rely on it to initiate extraterritorial litigation or reputational sanctions, demanding disclosure of training data composition and bias detection results—a lethal risk point easily overlooked in transnational compliance.
International Comparison
The EU and the U.S. display markedly different legal logics in their AI governance approaches. The EU, centered on the AI Act, adopts a hard-law regulatory model based on risk classification, categorizing AI applications into tiers such as “unacceptable risk”, “high risk”, and “limited risk”, and imposes mandatory data governance, transparency, human oversight, and conformity assessment obligations on high-risk systems, with violators facing fines of up to 7% of global annual revenue. For cross-border data flows, the EU treats personal data protection as a fundamental right through mechanisms such as adequacy decisions and standard contractual clauses under the General Data Protection Regulation [ https://eur-lex.europa.eu/eli/reg/2016/679/oj ], establishing stringent rules for cross-border transfers.
The United States has yet to enact comprehensive federal AI legislation, instead stitching together a governance network oriented toward innovation primacy and soft-law guidance through executive orders, voluntary commitments, and sectoral enforcement. Federal-level documents such as the Blueprint for an AI Bill of Rights and the NIST AI Risk Management Framework [ https://www.nist.gov/itl/ai-risk-management-framework ] are non-binding guidance, focused primarily on avoiding excessive regulation that could undermine technological leadership. On cross-border data, the U.S. has long advocated the free flow of commercial data, while simultaneously erecting technical barriers through export controls and outbound investment screening. Legal liability largely relies on traditional ex post remedies such as anti-discrimination and product liability law rather than ex ante approval.
In sum, the EU is rules-forward and rights-based; its provisions are vertical and status-conferring—the law tells an AI system “who you are” in advance and assigns corresponding obligations. The U.S. approach essentially sets its baseline at “unacceptable backwardness” rather than “unacceptable risk”, with compliance primarily embodied in industry-driven, flexible standards; its provisions are horizontal and outcome-oriented—the law typically adjudicates “what you caused” after harm occurs through existing sectoral statutes. If the EU model requires an AI system to hold a “passport of fairness and security”, and the U.S. model merely conducts a “judicial examination” after harm, the path of China’s Action Plan seeks to establish a network of “technical-legal bilateral mutual recognition agreements” for global AI trade and cooperation. It does not issue passports but provides the standard components and mutual recognition interfaces needed to build various types of passports. WAICO’s establishment provides an institutionalized assembly plant and delivery platform for these “standard components”.
Conclusion
The Action Plan is a programmatic document grounded in the legal bases of the “right to development” and “global public goods”, aimed at reshaping the international AI governance landscape. Its core lies in constructing, through its eight actions, a third path distinct from the U.S. approach of “innovation primacy and deregulation” and the EU approach of “risk classification and rights protection”. It provides an institutional platform for China’s transition from a “participant in AI governance” to a “rule-maker”. The establishment of WAICO and the release of the Action Plan together form a coupled structure of “organizational law plus policy catalogue”. Whether the organization can develop effective mechanisms in standard-setting, dispute resolution, and technical assistance will serve as a critical observation window for testing whether China can truly achieve a transformational shift in its role.
World Artificial Intelligence Cooperation Organization (WAICO)
On 16 July 2026, one day prior to the release of the Action Plan, 29 countries signed the Agreement on the Establishment of the World Artificial Intelligence Cooperation Organization in Shanghai, formally establishing WAICO with its permanent headquarters in Shanghai. This is the world’s first intergovernmental international organization dedicated exclusively to artificial intelligence cooperation. Its founding member states include China, Russia, Kazakhstan, Pakistan, Indonesia, Brazil, Venezuela, Laos, 10 African countries, and 12 Asian countries, among others. UN Secretary-General António Guterres attended the signing ceremony.
The establishment of WAICO carries the following significant implications:
1. A Breakthrough in the Law of International Organizations: WAICO was established by an “agreement” rather than a “treaty”, thereby lowering the threshold for domestic ratification by member states and reflecting a pragmatic design that facilitates participation by developing countries. Its organizational charter—including provisions on legal personality, privileges and immunities, and voting mechanisms—will serve as a new subject of study in the law of international organizations.
2. Institutional Empowerment of the Action Plan: The Action Plan is a policy document unilaterally released by China at the WAIC. Its specific commitments—such as providing 5,000 AI training slots to developing countries over the next five years and deploying the “Mazu” meteorological early warning system in 30 countries—are closely aligned with WAICO’s functions and may be viewed as the “start-up funding” and “initial projects” contributed by China to WAICO. This elevates the Action Plan from a unilateral policy declaration to an implementation commitment linked to a multilateral organizational framework.
3. Institutionalization of Rule Competition: While the EU projects its rules through the AI Act [ https://eur-lex.europa.eu/eli/reg/2024/1689/oj ] and the United States exerts pressure through bilateral digital trade agreements and export controls, WAICO is the first dedicated international AI organization initiated by China and headquartered in a Chinese city. It marks a shift in China’s AI governance pathway from “proposing rules” to “institutionalized rule export.”
Notably, India has not joined WAICO. This may affect the organization’s representativeness in coordinating AI governance in the Asia-Pacific region and leaves a variable that requires separate handling in the compliance planning of Chinese enterprises in that market.
The Action Plan
The Action Plan is anchored in the principle that “artificial intelligence should be an international public good that benefits all humanity.” It addresses data, computing power, ecosystems, empowerment, talent, rules, governance, and ethics, setting forth eight actions: Quality Data Supply, Inclusive Intelligent Computing Power, Open-Source Ecosystem Sharing, Deep AI Empowerment, Joint Cultivation of Digital and AI Talents, Joint Development of Rules and Standards, Security Governance Collaboration, and AI for Good. These actions respond to United Nations initiatives to strengthen international AI cooperation, bridge the digital divide, and leverage AI for sustainable development. The most legally impactful actions are outlined below.
1. Quality Data Supply: This action proposes to “promote cross-border data flows, build and operate cross-border trusted data spaces in certain sectors, and facilitate efficient, convenient, and secure cross-border data flows. It also aims to collaboratively construct high-quality corpora and industry-grade datasets, promote multilingual corpus co-construction and sharing, and lay a solid foundation for global AI innovation.” The “cross-border trusted data spaces” represents a “third way” distinct from the EU’s adequacy decisions and standard contractual clauses, and the U.S. model of free data flows. Its legal significance lies in deploying technical controls and contractual trust on a “space” basis, potentially giving rise to a new type of “special regulatory sandbox” for cross-border data. More bluntly, the “whereas” clauses and compliance schedules of transnational data contracts will soon need to answer whether, and how, to connect to such spatial architectures advocated by China. This is highly likely to become a focal point of negotiation in the governing law and dispute resolution clauses of data cooperation agreements.
2. Open-Source Ecosystem Sharing: This action proposes to “encourage the joint development of an international AI open-source community, carry out international exchanges and cooperation among open-source communities, and promote the sharing of general-purpose large models, foundational algorithms, and tool components. It calls for the collaborative development of open-source compliance systems and security guidelines, supports localized innovation by various countries based on open-source models, and seeks to build an open, shared, secure, orderly, and collaboratively governed global AI open-source ecosystem.” The phrase “collaboratively develop open-source compliance systems and security guidelines” goes beyond the scope of traditional open-source licenses by juxtaposing “security” and “compliance”. This signals the emergence of a “responsible open-source” standard that transcends traditional knowledge-sharing rules such as the Apache and GPL licenses, meaning that in the open-source model supply chain, security review and compliance commitments may soon shift from being optional to becoming prerequisites for market access.
3. Deep AI Empowerment: This action proposes to “deepen ‘AI+’ cooperation, establish transnational industry cooperation platforms, support digital and AI capacity building in developing countries, promote the regulated application and innovative development of agents, and advance AI application and empowerment in fields such as science, manufacturing, healthcare, education, agriculture, and governance, so as to use AI to promote economic development, improve social governance, and enhance people’s well-being.” By expressly mentioning “promote the regulated application and innovative development of agents”, this is the first time “agents” are given independent expression in a national-level international cooperation document. Its significance is that agents are moving from being mere tools to objects of legal concern with autonomous decision-making capabilities. This essentially opens the first window for international coordination on the legal vacuum of “agent liability”. As agents enter fields such as manufacturing, healthcare, and governance, their “regulated application” may evolve into mandatory standards covering algorithm filing, behavioral boundaries, intervention rights, and termination mechanisms, directly impacting product liability and professional ethics rules.
4. Joint Development of Rules and Standards: This action proposes to “jointly build AI standards and specification systems, promote the development and revision of international standards, collaboratively construct standard alignment and coordination mechanisms, and foster inclusiveness and interoperability among standard systems. It also calls for strengthening the alignment and coordination of AI development strategies, governance rules, and technical standards.” This is the part of the Action Plan with the greatest legal leverage. It seeks to replicate the successful experience of the telecommunications sector by embedding Chinese-led technical and security standards into global AI supply chain compliance through a “standards-essential” pathway.
5. Security Governance Collaboration: This action proposes to “jointly build AI security governance mechanisms, strengthen cybersecurity threat information sharing and emergency response cooperation, and prevent the misuse and abuse of AI technologies. It promotes research to enhance AI explainability, transparency, and security, strengthens AI data governance, and fosters a sound environment for global AI development. It advocates for establishing open platforms to share best practices and promote international cooperation on AI security governance globally.” Once “enhancing AI explainability and transparency” is incorporated into international cooperation mechanisms, it will directly translate into evidentiary rule challenges in cross-border litigation. In cases involving algorithmic discrimination or harm from automated decision-making, plaintiffs may rely on this to assert that defendants bear an “explainability obligation,” with the standard of that obligation referencing Chinese-led international security guidelines. This would elevate “explainability” from a technical ethics term to a legal standard of due diligence, triggering disputes over the cross-border discovery and admissibility of novel forms of evidence such as algorithm audit reports and model cards.
6. AI for Good: This action proposes to “uphold the AI scientific and technological ethical principles of openness and transparency, privacy and security protection, and controllability and trustworthiness, and jointly build a system of AI ethical guidelines. It implements the people-centered and AI-for-good philosophy to promote the building of a warmer intelligent society. It aims to eliminate racial discrimination, other forms of discrimination, and algorithmic bias, and to safeguard fairness and non-discrimination. It advances international governance cooperation on AI scientific research, contributes public science products for the Global South, and serves the UN 2030 Sustainable Development Goals.” By proposing to “eliminate racial discrimination, other forms of discrimination, and algorithmic bias, and safeguard fairness and non-discrimination” while contributing public goods for the “Global South,” this action aligns itself with both the UN 2030 Agenda and international human rights law discourse. It may become a “soft mandatory standard” against which the compliance of AI products exported by Chinese enterprises is judged. NGOs and competing jurisdictions may rely on it to initiate extraterritorial litigation or reputational sanctions, demanding disclosure of training data composition and bias detection results—a lethal risk point easily overlooked in transnational compliance.
International Comparison
The EU and the U.S. display markedly different legal logics in their AI governance approaches. The EU, centered on the AI Act, adopts a hard-law regulatory model based on risk classification, categorizing AI applications into tiers such as “unacceptable risk”, “high risk”, and “limited risk”, and imposes mandatory data governance, transparency, human oversight, and conformity assessment obligations on high-risk systems, with violators facing fines of up to 7% of global annual revenue. For cross-border data flows, the EU treats personal data protection as a fundamental right through mechanisms such as adequacy decisions and standard contractual clauses under the General Data Protection Regulation [ https://eur-lex.europa.eu/eli/reg/2016/679/oj ], establishing stringent rules for cross-border transfers.
The United States has yet to enact comprehensive federal AI legislation, instead stitching together a governance network oriented toward innovation primacy and soft-law guidance through executive orders, voluntary commitments, and sectoral enforcement. Federal-level documents such as the Blueprint for an AI Bill of Rights and the NIST AI Risk Management Framework [ https://www.nist.gov/itl/ai-risk-management-framework ] are non-binding guidance, focused primarily on avoiding excessive regulation that could undermine technological leadership. On cross-border data, the U.S. has long advocated the free flow of commercial data, while simultaneously erecting technical barriers through export controls and outbound investment screening. Legal liability largely relies on traditional ex post remedies such as anti-discrimination and product liability law rather than ex ante approval.
In sum, the EU is rules-forward and rights-based; its provisions are vertical and status-conferring—the law tells an AI system “who you are” in advance and assigns corresponding obligations. The U.S. approach essentially sets its baseline at “unacceptable backwardness” rather than “unacceptable risk”, with compliance primarily embodied in industry-driven, flexible standards; its provisions are horizontal and outcome-oriented—the law typically adjudicates “what you caused” after harm occurs through existing sectoral statutes. If the EU model requires an AI system to hold a “passport of fairness and security”, and the U.S. model merely conducts a “judicial examination” after harm, the path of China’s Action Plan seeks to establish a network of “technical-legal bilateral mutual recognition agreements” for global AI trade and cooperation. It does not issue passports but provides the standard components and mutual recognition interfaces needed to build various types of passports. WAICO’s establishment provides an institutionalized assembly plant and delivery platform for these “standard components”.
Conclusion
The Action Plan is a programmatic document grounded in the legal bases of the “right to development” and “global public goods”, aimed at reshaping the international AI governance landscape. Its core lies in constructing, through its eight actions, a third path distinct from the U.S. approach of “innovation primacy and deregulation” and the EU approach of “risk classification and rights protection”. It provides an institutional platform for China’s transition from a “participant in AI governance” to a “rule-maker”. The establishment of WAICO and the release of the Action Plan together form a coupled structure of “organizational law plus policy catalogue”. Whether the organization can develop effective mechanisms in standard-setting, dispute resolution, and technical assistance will serve as a critical observation window for testing whether China can truly achieve a transformational shift in its role.