Artificial intelligence is reshaping the global landscape of dispute resolution with unprecedented depth, quietly permeating the entire lifecycle of dispute resolution—from legal research and document drafting to evidence analysis and decision‑making support. As a key player in the international dispute‑resolution arena, China’s arbitration community stands at the cusp of an efficiency revolution driven by technological advancement, yet it also confronts multiple risks and challenges, including algorithmic black boxes, data breaches, and the implicit surrender of adjudicative authority. In the face of this historic transformation, guided by national-level policy frameworks such as the CPC Central Committee’s “Suggestions on Formulating the 15th Five-Year Plan for National Economic and Social Development,” the “New Generation Artificial Intelligence Development Plan,” and the “Global AI Governance Initiative,” the “Normative Guidelines on the Use of Artificial Intelligence in Arbitration” (a recommended text) — hereinafter referred to as the “Guidelines” — adopts an academic perspective to articulate a code of conduct that aligns with the realities of China’s arbitration development, embraces international best practices, and steers technological progress toward socially beneficial outcomes. By adopting the form of a “recommended text,” the Guidelines seek to fully leverage their role as a policy‑oriented think tank on critical issues of international law, providing intellectual support for national governance of emerging technologies, building a foundation of empirical evidence for future AI‑related legislation, and offering the arbitration sector a benchmark for soft‑law regulation. At the same time, we are keenly aware that the refinement of these Guidelines depends on rigorous testing and calibration through arbitration practice, requiring the concerted participation of both academia and the legal profession to achieve dynamic optimization. Only in this way can we strike a precise balance between technological innovation and procedural fairness, chart a path of smart arbitration that is uniquely Chinese yet globally leading, and inject sustained momentum into the enhancement of China’s arbitration credibility and its internationalization.

In recent years, the groundbreaking advances in generative artificial intelligence have sparked an explosive surge in its applications within the legal field. In arbitration practice, arbitral secretaries leverage AI technologies to generate procedural documents and conduct legal research; arbitrators rely on AI‑powered tools to receive case‑law recommendations and verify reasoning behind awards; and parties use AI systems to analyze vast volumes of evidentiary materials—scenarios that are no longer confined to theoretical discourse. However, the dual nature of technology is equally pronounced: algorithmic bias can entrench or even amplify social discrimination; the geographically limited scope of training data may lead to distortions in the application of law; and if unvetted AI‑generated content is directly adopted as a basis for awards, it could undermine the exclusive and solemn character of judicial authority. Moreover, the disclosure of sensitive arbitral data to public platforms risks triggering systemic breaches of confidentiality. Current legal frameworks either lack targeted regulations addressing these issues or confine such provisions to broader legislation on cybersecurity, personal data protection, and the governance of generative AI, making it difficult to meet the specialized, confidential, and final‑and‑binding requirements of arbitration.

Against this backdrop, the formulation of this Code of Practice carries threefold significance: first, it provides stakeholders in arbitration with clear expectations for conduct, thereby reducing compliance costs and ethical concerns associated with the adoption of new technologies; second, it seeks to strike a dynamic balance among supporting the use of emerging technologies to facilitate arbitration, upholding the fairness of arbitral proceedings, and safeguarding the legitimate rights and interests of the parties, thus preventing the misuse of technology from eroding public confidence in arbitration; and third, it articulates the Chinese arbitration community’s stance—open and inclusive yet prudent and rational—on technological innovation, offering Chinese insights to inform the development of international rules.

This “Guidance on Standards” draws upon and incorporates cutting-edge research and developments in the field of dispute resolution, both domestically and internationally. At the judicial level, China’s Supreme People’s Court, in its 2022 “Opinions on Regulating and Strengthening the Judicial Application of Artificial Intelligence,” was among the first to establish fundamental principles such as security and legality, as well as fairness and impartiality. Similarly, in 2024, Singapore’s Supreme Court and Canada’s Federal Court issued guidelines on the use of AI in litigation, setting forth clear requirements regarding transparency, data protection, and accountability. In the arbitration sphere, the Chartered Institute of Arbitrators (CIArb) in the United Kingdom, the American Arbitration Association (AAA), the Silicon Valley Arbitration and Mediation Center (SVAMC), and the Stockholm Chamber of Commerce Arbitration Institute (… SCC ), the Vienna International Arbitral Centre (VIAC), as well as institutions such as the China International Economic and Trade Arbitration Commission (CIETAC) and the Guangzhou Arbitration Commission (GZAC), have also, in recent years, issued guidelines or recommendations on AI‑assisted arbitration.

Despite differences in their historical contexts and forms of expression, a review of current global texts and practices reveals several key consensuses emerging in the application of artificial intelligence to human‑centered dispute resolution. These include: afofficeing the irreplaceable role of humans in the decision‑making process and the ultimate accountability of human actors; requiring users to fully recognize the limitations of the technology and to establish appropriate review and validation mechanisms; encouraging adequate disclosure of AI‑related practices to enhance transparency and credibility; and vigilantly guarding against algorithmic bias that may infringe upon the rights and interests of the parties involved, thereby upholding fairness and impartiality in both procedures and outcomes.

Building on existing consensus, this Code of Practice underscores that the use of artificial intelligence in arbitration must adhere to the fundamental principles of “risk prevention,” “self‑responsibility,” “fairness and impartiality,” “confidentiality,” and “transparency and trustworthiness.” Accordingly, it establishes a corresponding framework of obligations: the duty of due diligence requires users to proactively identify technological risks and exercise reasonable care; the obligation of review and verification mandates a substantive assessment of AI‑generated outputs to mitigate overreliance on technology; the disclosure obligation encourages all parties to the arbitration to transparently disclose their use of AI; and the labeling obligation ensures that AI‑generated content remains identifiable.

While drawing on established domestic and international practices, this Regulatory Guidance has been creatively developed in light of the practical realities of Chinese arbitration, with its innovations concentrated across five key dimensions.

The “Guidelines” articulate the principle of “encouraging use,” thereby signaling support for the application of artificial intelligence in arbitration, as opposed to a stance of cautious conservatism or technological neutrality. The aim of this “encouragement” is to enable all parties involved in arbitration to leverage AI technologies and tools in appropriate ways, thus enhancing both the efficiency and quality of dispute resolution. In terms of concrete institutional design, this “encouragement” goes beyond mere principled declarations; it is integrated into differentiated regulatory frameworks governing the conduct of various stakeholders. Article 13 expressly permits arbitral institutions to “actively utilize” AI across a range of scenarios, including but not limited to advisory services, case filing and acceptance, and review of awards. Article 23 grants arbitral tribunals the authority to “reasonably employ” AI, with the goal of improving procedural efficiency and the quality of awards. Meanwhile, Article 29 allows parties and their representatives to “agree upon the use” of AI, fully reflecting the principle of party autonomy in arbitration. This development‑oriented normative approach seeks to guide the arbitration community in proactively adapting to technological change, while also addressing potential concerns through the establishment of clear behavioral standards, thereby providing institutional impetus for the digital and intelligent transformation of the arbitration field.

Existing guidelines and frameworks on the use of artificial intelligence in arbitration largely focus on the activities of arbitral tribunals, parties, and their representatives. By contrast, this “Standardized Guidance” takes a pioneering step by dedicating an independent chapter to the development, introduction, and deployment of AI by arbitral institutions, marking a significant breakthrough in the scope of regulatory coverage. This innovation stems from an in-depth examination of institutional arbitration practice in China: institutions not only serve as organizers of arbitral services but also play a leading role in the adoption of new technologies and bear primary responsibility for risk management. Article 15 grants institutions the autonomy to develop or introduce AI, while also permitting them, when necessary, to opt for localized or private‑cloud deployments, thereby ensuring the security of core data and enabling flexible control over technology‑related costs. Article 16 clarifies institutions’ obligations to disclose their use of AI and to exercise oversight, encouraging them to evolve from mere users into service providers. Moreover, Article 22 groundbreakingly imposes on institutions a duty to exercise reasonable supervision over other arbitration participants and third‑party vendors. This institutional design places arbitral institutions at the forefront of technological governance, positioning them both as beneficiaries of AI applications and as gatekeepers of risk prevention, thus filling a critical gap in traditional arbitration rules regarding the regulation of institutional responsibilities in the realm of technology.

As international arbitration cases become increasingly complex and diverse, arbitral tribunals employ Arbitral Tribunal Secretary The use of auxiliary procedures to facilitate proceedings is becoming increasingly common, and Chinese arbitration practice relies heavily on the work of institutional arbitration secretaries. However, existing guidelines and handbooks in the field of arbitration have failed to provide regulatory frameworks governing the use of artificial intelligence by secretaries. Article 6 of this “Regulatory Guidance” distinguishes between “arbitral tribunal secretaries” and “…”. Case Management Secretary and provides “dual-track” regulations in Articles 21 and 28, thereby effectively addressing the differentiated arbitration secretarial systems in China and abroad. As an institutional representative, the case‑management secretary’s use of artificial intelligence is brought within the institution’s unified compliance framework; by contrast, the tribunal secretary, as an extension of the arbitral tribunal, is subject to the same rigorous review obligations as the tribunal itself. This categorical approach both respects the differences between the two in their appointment procedures and functional roles, while ensuring a clear and traceable chain of accountability, thus reflecting the organic integration and coordinated regulation of institutional administrative functions and tribunal adjudicative functions within the context of AI governance.

Most existing guidelines or frameworks for the use of artificial intelligence in arbitration require arbitral tribunals to review and verify AI-generated content, yet these provisions are often vague, failing to specify concrete verification procedures. This “Standardized Guidance” refines the tribunal’s obligations to review and verify: Paragraph 1 of Article 26 mandates that such reviews be conducted on the basis of “original materials, legal texts, and other reliable sources,” while Paragraph 2 expressly prohibits the circular validation practice of “using AI to validate AI.” This dual‑pronged approach both establishes the tribunal’s required course of action and delineates red lines—clearly defining what is prohibited—while articulating specific methods for review and verification. The aim of this institutional design is to address the “echo chamber effect” that may arise from the homogeneity of training data used by large language models, as well as to counteract the tendency toward “technologically driven complacency” that can emerge in practice. By refining the review and verification process, it seeks to ensure meaningful human oversight, prevent AI from becoming a “judge of judges,” and safeguard the exclusive and human‑centric nature of arbitral adjudication.

This Code of Practice places the protection of parties’ rights at the heart of technological governance. Article 29 grants parties the right to agree on the use of artificial intelligence, thereby respecting their autonomy of will; Article 33 encourages proactive disclosure by the parties, fostering information symmetry; Article 32 imposes a requirement to submit AI-generated evidence accompanied by “clear labeling plus usage records,” thus safeguarding the opposing party’s right to challenge such evidence; and Article 34 goes further by establishing, for the first time, remedial measures, stipulating that when the use of AI may substantially harm a party’s rights and interests, that party may request the arbitral tribunal or the relevant institution to review the matter and take appropriate steps. This comprehensive chain of rights protection—spanning “ex ante agreements,” “in‑process oversight,” and “ex post remedies”—complements the provisions of Article 27, which encourage disclosure and communication and consultation, thereby forming a closed-loop system that balances rights, obligations, and responsibilities.

In order to promote the application of artificial intelligence in arbitration, regulate its use, mitigate technological risks, strike a balance between enhanced efficiency and procedural fairness, safeguard the legitimate rights and interests of the parties, and enhance the credibility of arbitration, this Guideline is hereby formulated.

1. This article sets forth the purpose of formulating these normative guidelines.

2. This guideline is designed to proactively address the rapid advancement of artificial intelligence technologies, particularly generative AI, by supporting their appropriate application in arbitral proceedings, while also giving due consideration to the potential risks associated with these technologies, including but not limited to algorithmic bias, algorithmic opacity, false generation (“hallucinations”), data security and information leakage, as well as other issues that could adversely affect the credibility of arbitration.

3. This set of guidelines aims to provide arbitration participants with a clear and pragmatic framework for risk prevention and conduct. At its core, it seeks to promote the use of artificial intelligence in the arbitration field through standardized guidance, thereby enhancing arbitration efficiency and procedural fairness, safeguarding the legitimate rights and interests of the parties, and ultimately bolstering the overall credibility of arbitration.

This guideline may be applied to the use of artificial intelligence in institutional arbitration and ad hoc arbitration.

1. This Code of Practice applies to the use of artificial intelligence in institutional arbitration and ad hoc arbitration.

2. There are two fundamental modes of arbitration: institutional arbitration and ad hoc arbitration. Ad hoc arbitration can, in turn, be further categorized into scenarios with and without institutional support. Regardless of the mode, the arbitral tribunal, the parties, and their representatives all participate in the arbitral proceedings; accordingly, the provisions in this Code governing the use of artificial intelligence by the tribunal, the parties, and their representatives apply to all such cases. Arbitral institutions primarily engage in arbitral proceedings in the context of institutional arbitration and ad hoc arbitration with institutional assistance; therefore, the provisions in this Code concerning the use of artificial intelligence by arbitral institutions should be applied on a case-by-case basis, depending on the specific circumstances of their use.

An engineering system that generates outputs such as content, predictions, recommendations, or decisions in response to a given objective defined by humans.

1. The definition of artificial intelligence in this provision is based on the “Information Technology—Terminology for Artificial Intelligence” (GB/T 41867-2022), published by the State Administration for Market Regulation of China (National Standardization Management Committee). This standard provides separate definitions for “artificial intelligence” and “artificial intelligence system.” Specifically, Clause 3.1.2 defines “artificial intelligence” as “the research and development of mechanisms and applications related to artificial intelligence systems,” while Clause 3.1.8 defines “artificial intelligence system” as “an engineering system that, in response to a given objective defined by humans, can produce outputs such as content, predictions, recommendations, or decisions.” Furthermore, the European Union’s… Artificial Intelligence Act Article 3(1) of Regulation (No. 2024/1689) defines an “artificial intelligence system” as “a machine-based system designed to operate with varying degrees of autonomy and, once deployed, may exhibit adaptability; such a system infers how to generate outputs—such as predictions, content, recommendations, or decisions—based on the inputs it receives, in order to achieve explicit or implicit objectives, where these outputs can influence physical or virtual environments.” Given that this Guidance aims to regulate practical applications rather than define disciplinary concepts, artificial intelligence is defined here as a category of systems.

An artificial intelligence capable of generating text, images, audio, video, and other types of content, based on algorithms, models, and rules.

1. The definition of generative artificial intelligence in this provision draws upon Article 22, Paragraph 1 of the Interim Measures for the Administration of Generative Artificial Intelligence Services, jointly issued by the Cyberspace Administration of China and the National Development and Reform Commission, among other authorities, which defines “generative artificial intelligence technology” as “models and related technologies capable of generating content such as text, images, and audio.” Building on this definition, this provision further specifies the prerequisite that such technology is “based on algorithms, models, rules, and the like,” thereby treating generative artificial intelligence as a specific subtype of artificial intelligence and ensuring effective consistency with the preceding paragraph’s definition of artificial intelligence.

Any process by which an artificial intelligence system is calibrated to align with organizational objectives or ethical principles.

1. The definition of AI value alignment set forth in this provision draws upon Article 2 of the UK Courts and Tribunals Judiciary’s “Guidance on the Use of Artificial Intelligence for Judicial Personnel (2025).” According to that provision, AI value alignment is an ethically grounded concept aimed at ensuring that the goals, behaviors, and outcomes of artificial intelligence are consistent with the value systems widely recognized by human society, thereby guiding AI to align with prevailing societal values.

2. The core principles of AI value alignment can be summarized in three key points: first, aligning AI’s judgments with human value judgments; second, aligning AI’s actions with human value‑driven behaviors; and third, aligning AI’s goals with human value‑based objectives.

3. Within the context of this guideline, value alignment means that, in the course of deploying artificial intelligence, stakeholders must ensure its fairness, prevent bias, and avoid deviations from mainstream societal values or violations of public order, good morals, and the public interest.

Organizations and individuals participating in the resolution of arbitration disputes include arbitral institutions, arbitral tribunals and arbitrators, parties to the arbitration, legal representatives, secretaries of the arbitral tribunal, third parties to the arbitration, witnesses, experts, and others.

Arbitration secretaries include the tribunal secretary and the case‑management secretary. The tribunal secretary is a professional designated by the arbitral tribunal to provide assistance to the tribunal. The case‑management secretary is a professional appointed by the arbitral institution to administer and assist with its arbitration proceedings.

1. This provision adopts an open-ended enumeration to define the participants in arbitration, using “participation in dispute‑resolution proceedings” as the substantive criterion for determination. The scope encompasses both legal persons and non‑legal person organizations, as well as natural persons. The term “and the like” leaves room for interpretation, thereby covering entities that may participate in arbitration under other specific circumstances, such as expert witnesses, appraisers, translators, industry associations, and the like.

2. The classification of arbitration secretaries set forth in this Article is based on the “Standards for Arbitration Secretaries (Recommended Text)” jointly drafted and issued by the Institute of International Law of the Chinese Academy of Social Sciences and the Guangzhou Arbitration Commission, which divides arbitration secretaries into tribunal secretaries and case‑management secretaries, and accordingly regulates their use of artificial intelligence in arbitral proceedings.

Parties to arbitration are encouraged to make appropriate use of artificial intelligence–assisted arbitration to enhance both the efficiency and quality of the proceedings.

1. This provision innovatively proposes that “encouraging its use” should be established as a fundamental principle in the application of artificial intelligence in arbitration.

2. Globally, broad consensus has emerged around core principles governing AI‑assisted arbitration—such as human accountability, confidentiality, fairness and impartiality, and transparency—yet varying degrees of acceptance persist. For instance, the Vienna International Arbitral Centre’s “Guidelines on the Use of Artificial Intelligence in Arbitral Proceedings” adopts a cautious stance of “no provision, no promotion, no liability,” whereas the Stockholm Chamber of Commerce Arbitration Institute’s “Guidelines on the Use of Artificial Intelligence in Cases Administered under the SCC Rules” and the Chartered Institute of Arbitrators’ “Guidance on the Use of Artificial Intelligence in Arbitration (2025)” take a technologically neutral approach. By contrast, this Code explicitly embraces a supportive and encouraging position, aiming to foster the deep integration of AI into arbitral practice and thereby enhance the overall effectiveness of arbitration services. As a data‑driven technology, AI relies on trustworthy data drawn from real‑world arbitral proceedings to refine and upgrade its models, enabling continuous training and iteration. Only through such iterative processes can AI technologies and solutions better align with the specialized needs of arbitration, ultimately advancing digital justice to a higher standard.

3. This Guideline encourages arbitration participants to utilize artificial intelligence to support arbitration, but it does not mandate its uniform adoption. Instead, it allows participants to independently decide whether and how to employ AI‑assisted arbitration, taking into account the specific circumstances of each case, associated costs, and other relevant factors. In light of the distinct roles and characteristics of different arbitration stakeholders, the Guideline provides differentiated provisions on such encouragement: arbitral institutions may “actively leverage” AI (Article 13); arbitral tribunals may “appropriately employ” AI (Article 23); and parties and their representatives may “agree to use” AI (Article 29).

Arbitration participants should be familiar with the basic functions and potential risks of the artificial intelligence systems they employ, and use AI‑assisted arbitration in a prudent and responsible manner.

Arbitration participants shall review and verify AI-generated content and may not rely directly on such content,未经复核验证, as the basis for their independent decision-making, in order to mitigate the technological risks associated with artificial intelligence.

1. This provision establishes principles for risk mitigation in the use of artificial intelligence in arbitral proceedings, clarifying the auxiliary role of AI tools and imposing on all parties to the arbitration an obligation to review and verify any content generated by such tools. Given the inherent limitations of AI technology, which may produce erroneous information or false content, reliance on unverified outputs could undermine the accuracy and impartiality of the arbitral process.

2. The first paragraph requires that arbitration participants be familiar with the basic functions and potential risks of the artificial intelligence they use. First, they should have a clear understanding of the tasks AI can perform and the scenarios in which it is applicable; second, they should be aware of the risks associated with its deployment. For example, Clause 3.2 of the Chartered Institute of Arbitrators’ “Guidance on the Use of Artificial Intelligence in Arbitration (2025)” states: “Parties and arbitrators are encouraged to gain a thorough understanding of the various risks involved in any potential applications of artificial intelligence, and to weigh the benefits of using AI tools against the arbitration‑related risks arising from their deployment.”

3. Paragraph 2 clarifies the obligation of arbitration participants to review and verify AI-generated content. Specifically, arbitral institutions, arbitral tribunals, parties to the arbitration, and their representatives are required, when using artificial intelligence, to independently review and verify the content it generates; unverified AI-generated content may not be directly relied upon as the basis for fact-finding, legal application, or decision-making in arbitral proceedings. For example, Article 13 of the “Guidelines on the Use of Generative Artificial Intelligence by Judges, Judicial Officers, and Support Staff of the Judiciary of Hong Kong (2024)” stipulates: “Any biased content in the training data—whether cultural or moral—as well as regional biases or misinformation will inevitably be reflected in the generated responses. Judges, judicial officers, and support staff should bear this in mind and make any necessary corrections before using or relying on such generated information.”

4. The scope of review and verification covers all content generated by artificial intelligence, including but not limited to outputs in the form of various texts, data, analytical conclusions, and other formats. The review and verification shall focus on assessing the authenticity, accuracy, completeness, and appropriateness of the content.

Arbitration participants shall bear responsibility for any documents they submit that contain content generated by artificial intelligence.

Arbitration participants shall adhere to the principle of AI-assisted decision-making and may not rely exclusively on artificial intelligence or allow it to substitute for human decision-making.

1. Paragraph 1 of this article establishes the principle of individual liability, clarifying that no participant in arbitration may invoke the use of artificial intelligence as a basis for excluding or mitigating their own responsibility. Whether it be an arbitral institution, an arbitral tribunal, an arbitration agent, or any other participant, each shall bear independent responsibility for the documents they issue, including the statement of claim, the statement of defence, procedural orders, and arbitral awards. For example, Article 4.2 of the Vienna International Arbitration Centre’s “Guidelines on the Use of Artificial Intelligence in Arbitral Proceedings” expressly provides: “Arbitrators and parties shall be liable for the outputs of any artificial intelligence tools they employ.”

2. Paragraph 2 of this article clarifies that artificial intelligence may serve only as an auxiliary tool and must not be used to replace human decision-making. For example, the arbitral tribunal is required to review, assess, and verify the results generated by AI, and it is prohibited to allow AI to substitute for the tribunal in rendering an award, thereby safeguarding the authority and credibility of arbitral awards. As stipulated in Article 3.2 of the “Guidelines on the Use of Artificial Intelligence Technologies in Arbitration (Trial)” issued by the China International Economic and Trade Arbitration Commission: “Arbitral awards shall be rendered by the arbitral tribunal. The development of artificial intelligence technologies shall not supplant the arbitrators’ duties in adjudicating cases. Any outputs produced by AI tools may serve solely as auxiliary support for the tribunal’s deliberations, and the tribunal shall bear responsibility for the arbitral award rendered with the assistance of such AI tools.”

The use of artificial intelligence by arbitration participants shall be fair and impartial, and measures shall be taken to prevent AI-generated content from adversely affecting the legitimate rights and interests of any party due to factors such as training data, algorithm design, or application scenarios.

1. This provision establishes the principles of fairness and impartiality that artificial intelligence must adhere to in arbitration proceedings.

2. This principle is grounded in the potential bias risks associated with the application of artificial intelligence technologies: if training data contain historical or societal biases related to gender, geography, culture, or other dimensions, or if the algorithm design itself harbors implicit discrimination, the resulting outputs may, in fact‑based analyses, linguistic formulations, or recommended conclusions, systematically disadvantage a particular party in an arbitration proceeding. Moreover, given the substantial variations among arbitration cases in terms of the type of dispute, industry context, and applicable law, general‑purpose AI models that have not undergone targeted fairness calibration for arbitration settings will struggle to impartially resolve disputes arising in specialized or context‑specific scenarios. Furthermore, the use of technology tools with bias risks in arbitral proceedings without appropriate regulatory safeguards not only constitutes a procedural defect but may also undermine the validity and enforceability of the arbitral award.

3. This principle represents an extension and refinement of the core concept of fair and impartial dispute resolution as applied to AI‑driven contexts. For example, Article 4 of the Supreme People’s Court of China’s “Opinions on Regulating and Strengthening the Judicial Application of Artificial Intelligence” (Fa Fa [2022] No. 33) explicitly establishes the “principle of fairness and impartiality,” mandating that “judicial principles be upheld and impartial justice be served, ensuring that AI products and services are free from discrimination and bias, and that the integrity of judicial proceedings and outcomes is not compromised by technological intervention or biases in data or models.”

Parties to arbitration who utilize AI‑assisted arbitration shall comply with national laws and regulations on cybersecurity, data security, and the protection of personal information, and shall adhere to the principle of confidentiality in arbitration.

Arbitration participants shall not input any arbitration-related information involving state secrets, commercial secrets, or sensitive personal data into unauthorized public artificial intelligence platforms or tools that lack adequate security safeguards.

1. This provision aims to underscore the principle of confidentiality that must be observed when applying artificial intelligence in arbitration. Confidentiality is a fundamental feature of the arbitration system and one of the key reasons parties choose arbitration as a dispute‑resolution mechanism; its core lies in safeguarding trade secrets and sensitive personal information. The use of AI often entails data transmission, storage, and processing, which may objectively increase the risk of information leakage. Accordingly, when employing AI, it is imperative to strictly comply with the relevant provisions of the Arbitration Law, the Cybersecurity Law, the Data Security Law, and the Personal Information Protection Law. For example, Article 5(6) of the Singapore Supreme Court’s “Guidelines on the Use of Generative Artificial Intelligence Tools by Court Users (2024)” stipulates: “When using generative AI tools, you must adhere to confidentiality orders, the Personal Data Protection Act, intellectual property laws, and applicable legal privilege rules.”

2. This provision imposes a dual restriction on the selection of artificial intelligence platforms or tools: it prohibits the input of arbitration‑related information involving state secrets, commercial secrets, or sensitive personal data into public AI platforms or tools that lack proper authorization or adequate security safeguards. Public AI platforms or tools operating without lawful authorization typically fail to undergo compliance reviews in their data collection, model training, and content generation processes, thereby posing risks of unlawful use. Moreover, such platforms—being accessible to an open user base—may exploit user‑provided data for model optimization and output, potentially resulting in the recording, retention, or transmission of that data to third parties. Consequently, the use of these public platforms or tools must be subject to reasonable restrictions to safeguard the confidentiality of arbitration proceedings. Similarly, Article 10 of the “Guidelines for the Use of Generative Artificial Intelligence by Judges, Judicial Officers, and Support Staff of the Judiciary of Hong Kong (2024)” restricts the use of publicly available AI tools: “Judges, judicial officers, and support staff shall not enter any private, confidential, or sensitive information into open or publicly accessible generative AI chatbots; they should also ensure that the input is sufficiently broad and that any identifying details have been anonymized. Where the chatbot offers a chat‑history function, that feature should be disabled.” Given the rapid pace of AI development, the wording of this provision could be further refined to more comprehensively and proactively address relevant AI platforms and tools.

Parties to arbitration are encouraged to employ artificial intelligence in a transparent and trustworthy manner, and, where necessary, to disclose information regarding its use, underlying functionalities, and intended purposes.

1. This provision encourages all participants in arbitration proceedings—including arbitral institutions, arbitral tribunals, parties, and their representatives—to proactively disclose the use of artificial intelligence. This measure aims to enhance the transparency of the arbitration process and ensure… Applications of Artificial Intelligence Technology Credibility.

2. The disclosed information must encompass three core elements: first, a “description of usage,” clearly specifying the stage and specific methods of the technology’s application; second, a “description of basic functions,” elucidating the technical characteristics and operating principles of the AI involved; and third, a “statement of purpose,” explicitly outlining the objectives that the technology is intended to achieve. For example, Article 6 of the Supreme People’s Court of China’s “Opinions on Regulating and Strengthening the Judicial Application of Artificial Intelligence” (Fa Fa [2022] No. 33), under the “Principle of Transparency and Trustworthiness,” stipulates: “When judicial AI products and services are deployed, they shall, in an easily understandable manner, disclose and label their respective functions, performance, and limitations, ensuring that the application process and outcomes are predictable, traceable, and trustworthy.”

3. Disclosure of the use of artificial intelligence is a concrete manifestation of the principle of integrity in arbitration. For example, the Stockholm Chamber of Commerce Arbitration Institute (SCC) explicitly states in the “Honesty” section of its “Guidelines on the Use of Artificial Intelligence in Proceedings under the SCC Rules”: “Arbitral tribunals are encouraged to disclose any use of artificial intelligence in their fact‑finding and legal analysis, as well as in the application of law to the facts.”

In both institutional arbitration and ad hoc arbitration, arbitral institutions can proactively leverage artificial intelligence to enhance the efficiency and quality of arbitration, encompassing, but not limited to, the following scenarios: advisory services, case filing and acceptance, tribunal composition, procedural management, award review, similar‑case analysis, and legal research.

1. This provision aims to encourage arbitration institutions to actively leverage the technical characteristics and advantages of artificial intelligence, using it as an effective tool to enhance the efficiency and quality of dispute resolution. In its “Opinions on Standardizing and Strengthening the Judicial Application of Artificial Intelligence” (Fa Fa [2022] No. 33), the Supreme People’s Court of China explicitly states: “We will accelerate the deep integration of artificial intelligence technologies with judicial adjudication and enforcement, litigation services, judicial administration, and social governance; standardize the application of AI in the judiciary; improve the practical effectiveness of AI‑driven judicial applications; and promote the modernization of the adjudication system and adjudicative capabilities.” For arbitration institutions, artificial intelligence offers significant advantages in managing arbitration cases. For example, the Silicon Valley Arbitration and Mediation Center’s “Guidelines on the Use of Artificial Intelligence in Arbitration” notes: “The introduction of AI into arbitral proceedings can enhance procedural efficiency.” Similarly, the Stockholm Chamber of Commerce Arbitration Institute (SCC) states in the preamble to its “Guidelines on the Use of Artificial Intelligence in Proceedings under the SCC Rules”: “The use of AI in dispute resolution holds substantial potential for reducing costs and improving efficiency.”

2. The provision lists several scenarios in which arbitration institutions may employ artificial intelligence in their proceedings. For example: using AI‑powered customer service to respond to inquiries about arbitration services; conducting preliminary reviews of the completeness of documents submitted by the parties; intelligently recommending shortlists of arbitrators; automating procedural scheduling and other workflows; automatically verifying the format and citation compliance of arbitral awards (without engaging in substantive adjudication); and performing case retrieval and comparative analysis, among others.

3. This provision clarifies that arbitral institutions may employ artificial intelligence in administered arbitration proceedings and may also utilize relevant technologies to assist in the handling of ad hoc arbitration cases.

Arbitral institutions shall not employ artificial intelligence in cases involving national security or significant public interests.

1. This provision specifies the circumstances under which arbitration institutions may not employ artificial intelligence in their proceedings. Arbitration institutions are prohibited from using AI—whether as a standalone tool or merely in a supporting capacity—in cases involving national security, significant public interests, or other matters of comparable sensitivity. Given the inherently sensitive nature of such cases, and in order to effectively mitigate risks, it is imperative to maintain human oversight and control throughout the entire arbitration process. The automatic decision‑making by AI must be strictly barred at all stages, including case filing and acceptance, tribunal composition, and procedural management, thereby ensuring that ultimate authority over case administration remains officely in human hands.

2. This provision draws on the prudent stance set forth in the Guangzhou Arbitration Commission’s “Position Paper on Digital and Intelligent Arbitration and Recommended Guidelines for the Application of Artificial Intelligence in Dispute Resolution.” For sensitive cases involving national security, social stability, or significant public interests, a stricter principle of human oversight must be upheld. Given the potential for uncontrollable risks that artificial intelligence may pose when handling matters pertaining to national security, social stability, and major public interests, this regulatory guideline, guided by the principle of risk minimization, expressly prohibits the use of AI in such cases. Furthermore, in its “Opinions on Regulating and Strengthening the Judicial Application of Artificial Intelligence” (Fa Fa [2022] No. 33), the Supreme People’s Court of China has emphasized that judicial AI products and services must be developed, deployed, and operated in compliance with the law, without compromising national security or infringing upon legitimate rights and interests, thereby ensuring the protection of state secrets, cybersecurity, data security, and personal information.

Arbitral institutions may independently develop or procure from third parties artificial intelligence systems to support their arbitration proceedings.

Where the arbitral institution deems it necessary, it may require the localization or private deployment of artificial intelligence.

1. Paragraph 1 of this article expressly provides that an arbitration institution may either independently develop and deploy artificial intelligence or entrust a third party to do so. In the former case, the arbitration institution serves both as the provider and the user of the AI system. In the latter case, the arbitration institution is solely a user of the AI system, not its provider. The legal liabilities borne by the provider and the user of AI services may differ under certain circumstances.

2. Paragraph 2 of this article expressly stipulates that arbitral institutions have the authority to adopt either on-premises or private‑cloud deployment, primarily for confidentiality reasons. In its “Guidelines on the Use of Artificial Intelligence Systems in Courts and Tribunals,” published in 2025, UNESCO recommends, “Adopt, deploy, and utilize AI systems that comply with international standards for information access, in order to safeguard confidential information and, by establishing safeguards against cyber threats, ensure that judicial bodies maintain continuous control over their data.” For arbitral institutions, different deployment approaches may entail varying costs; this Guidance does not prescribe a uniform standard. Whether deploying on‑premises or in a private cloud, the key lies in implementing encrypted storage for core computational functions and sensitive data, thereby enabling arbitral institutions to maintain effective and ongoing control over their data and preventing the unauthorized disclosure of sensitive personal information, trade secrets, and other protected materials during arbitral proceedings.

Where an arbitral institution independently develops AI tools to assist in arbitration and generate content, it shall, as a service provider, fulfill its obligation to provide clear labeling.

Where an arbitral institution engages a third party to provide AI‑assisted content‑generation services, the third party, as the service provider, shall be required to affix appropriate identification, and the arbitral institution shall bear supervisory responsibility.

1. This provision sets forth the obligations of arbitral institutions to comply with artificial intelligence labeling requirements.

2. Paragraph 1 sets forth the circumstances under which an arbitration institution independently develops and deploys artificial intelligence. When an arbitration institution independently researches, develops, and deploys an AI system for content generation, it itself qualifies as an AI service provider and is required to fulfill its legal obligation to label such content. According to the “Technical Specification for the Identification of Artificial Intelligence‑Generated Synthetic Content,” jointly issued by the Cyberspace Administration of China, the Ministry of Industry and Information Technology, and other authorities (effective September 1, 2025), online information service providers must, in accordance with the “Regulations on the Administration of Deep Synthesis of Internet Information Services,” carry out AI‑generated content identification measures, including both visible and implicit labeling.

3. Paragraph 2 addresses situations in which an arbitration institution commissions the development and deployment of artificial intelligence. When the arbitration institution entrusts a third party with the research, development, and deployment of an AI system, that third party acts as the AI service provider and directly assumes the obligation to affix identification marks, while the arbitration institution bears a supervisory responsibility.

Arbitral institutions shall review AI-generated content to ensure that such content complies with the requirements of public order and good morals.

1. This provision establishes the obligation of arbitral institutions to align their use of artificial intelligence with core values. Given the probabilistic, predictive nature of generative AI, its outputs inherently carry the risk of containing unlawful or inaccurate information—particularly when training data are insufficient or information support is inadequate. In such circumstances, large language models are highly prone to generating discriminatory content that reinforces structural biases across dimensions such as race, gender, and cultural identity. In response, since proposing a human-centered approach in its 2019 “Policy Framework for Artificial Intelligence,” the Organisation for Economic Co-operation and Development (OECD) has further strengthened, in its 2024 update, value‑based requirements pertaining to safety and risk management, governance of synthetic content, privacy protection, intellectual property rights, and environmental sustainability.

2. As artificial intelligence becomes increasingly integrated into the field of dispute resolution, the principle of value alignment has gained widespread acceptance due to its dual role in safeguarding both technological effectiveness and ethical standards. Article 7 of the Supreme People’s Court of China’s “Opinions on Regulating and Strengthening the Judicial Application of Artificial Intelligence” (Fa Fa [2022] No. 33) underscores the “principle of public order and good morals,” ensuring that the judicial application of AI does not contravene these principles, nor undermine social public interests and order, or violate societal public morality and ethics. In the realm of arbitration-based dispute resolution, arbitral institutions are likewise required to conduct prudent reviews of AI-generated outcomes and promptly rectify any value deviations, thereby preventing harm to the legitimate rights and interests of the parties involved.

When an arbitration institution uses artificial intelligence to generate content, it shall review and verify the AI’s output.

1. This provision clarifies the arbitration institution’s obligation to review and verify outcomes generated by artificial intelligence.

2. This regulation focuses on outcome‑based oversight, aiming to address errors in content generation arising from issues such as AI “hallucinations,” bias, discrimination, and information omissions. In response, the Chartered Institute of Arbitrators’ “Guidance on the Use of Artificial Intelligence in Arbitration (2025)” designates “independent verification” as an obligation of arbitrators. Similarly, the Silicon Valley Center for Mediation and Arbitration’s “Guidance on the Application of Artificial Intelligence in Arbitration” imposes on the parties a duty to review AI‑generated outputs used in preparing submissions and to verify their accuracy. This Code of Practice treats arbitral institutions as participants in the arbitration process; since they too will utilize AI‑generated content, they are likewise required to assume a responsibility for review and validation.

3. Whether in institutional arbitration or in ad hoc arbitration with institutional support, the arbitral institution serves as a key organizer of the arbitral proceedings; however, it bears only a duty of review and verification with respect to content generated by artificial intelligence that it itself employs, while the obligation to review and verify outcomes produced by other parties to the arbitration rests with those respective parties themselves.

Arbitral institutions shall, through the adoption of rules, standards, guidelines, and other appropriate measures, disclose the extent to which they employ artificial intelligence in arbitration proceedings, thereby enhancing transparency.

1. This provision governs the disclosure by arbitral institutions of their use of artificial intelligence in arbitral proceedings, with the aim of upholding the principles of transparency and trustworthiness. By making public the relevant guidelines for the application of such technologies, arbitral institutions pursue two primary objectives: first, to enable parties to the arbitration to access, through open channels, the rules governing the use of AI in arbitral processes, thereby ensuring that its deployment complies with established standards; and second, to facilitate public awareness and oversight of the activities undertaken by arbitral institutions in employing AI to support arbitration.

2. Across the globe, numerous arbitration institutions have issued guidelines, directives, and rules related to artificial intelligence, seeking to establish principles for its application in arbitral proceedings and provide practical guidance on appropriate conduct. For example, the Silicon Valley Arbitration and Mediation Center released its “Guidelines on the Use of Artificial Intelligence in Arbitration” in April 2024; in March 2025, the American Arbitration Association–International Centre for Dispute Resolution published the “Guidelines for Arbitrators on the Use of AI Tools”; in November, the AAA introduced the “Rules on AI‑Driven Arbitration”; the Stockholm Chamber of Commerce Arbitration Institute (SCC) issued, in October 2024, the “Guidelines on the Use of Artificial Intelligence in Proceedings Administered under the SCC Rules”; the China International Economic and Trade Arbitration Commission released, in July 2025, the “Guidance on the Use of Artificial Intelligence Technologies in Arbitration (Trial)” ; and in August 2025, the Guangzhou Arbitration Commission promulgated the “Position Paper of the Guangzhou Arbitration Commission on Digital‑Intelligent Arbitration and Recommended Guidelines for the Application of Artificial Intelligence in Dispute Resolution,” among others.

When artificial intelligence is employed in arbitration, arbitral institutions, in their capacities as personal information processors, data processors, and providers or users of AI services, shall fulfill their statutory obligations under relevant laws and regulations pertaining to personal information protection, data security, and the governance of generative AI services.

1. This provision expressly stipulates that arbitral institutions, in the course of employing artificial intelligence, must comply with their statutory obligations to protect personal information and ensure data security. The introduction of AI by arbitral institutions is intended to enhance the efficiency and quality of arbitration and to provide parties with a superior arbitration experience; however, this must not result in the leakage of personal information or pose risks to data security.

2. In AI‑driven application scenarios, arbitration institutions often assume multiple roles—such as personal information processors, data controllers, and users of AI services—and must therefore bear the corresponding legal liabilities. In China, when employing AI to support arbitration, such institutions are required to comply with the legal obligations set forth in the Personal Information Protection Law, the Data Security Law, the Provisional Measures for the Administration of Generative Artificial Intelligence Services, and other relevant regulations. Furthermore, the ICC Commission on Arbitration and ADR, in its 2022 publication “Harnessing Technology to Promote Fair, Effective, and Efficient International Arbitration Proceedings,” explicitly states that all participants in arbitration must fulfill their obligations related to cybersecurity vulnerabilities and data protection.

Case management secretaries may employ artificial intelligence in providing administrative and support services for arbitration proceedings, provided they possess a sound understanding of AI’s basic functionalities and potential risks, and conduct thorough review and verification of content generated by AI.

1. This provision aims to clarify the risk‑mitigation obligations of the case management secretary in AI‑assisted arbitration. As a core figure responsible for implementing case management within the arbitral institution, the case management secretary plays a pivotal role in ensuring the efficient conclusion of arbitral proceedings and serves as a key actor in the application of AI‑assisted arbitration; accordingly, their conduct must be subject to appropriate regulation.

2. The obligations of risk prevention primarily encompass two aspects: the “duty to be informed of risks” and the “duty to mitigate risks.” On the one hand, the case management secretary shall have a thorough understanding of the various risks that may arise from the use of artificial intelligence, including potential harm to the legitimate rights and interests of the parties to the arbitration, as well as adverse effects on the fairness of the arbitral proceedings. On the other hand, the case management secretary is required to review and verify the outputs generated by artificial intelligence in order to guard against technological risks.

Arbitral institutions shall, within a reasonable scope, supervise the use of artificial intelligence by arbitration participants in cases administered by such institutions.

When an arbitral institution introduces artificial intelligence from a third party to assist in the arbitration process, it shall, within reasonable limits, supervise the third-party provider’s compliance with applicable laws and regulations.

1. This article sets forth the provisions regarding the arbitration institution’s duties to supervise and manage the use of artificial intelligence in arbitration proceedings.

2. Paragraph 1 clarifies the scope of supervisory oversight over arbitration institutions. Such institutions shall supervise the use of artificial intelligence–assisted decision‑making, document generation, or information processing by all participants in institutional arbitration proceedings, including arbitral tribunals, parties and their representatives, and arbitration secretaries. Through this oversight, it is possible to prevent the misuse of technology arising from issues such as algorithmic bias, implicit discrimination, or data‑security risks, thereby upholding both substantive and procedural fairness in arbitration and safeguarding its credibility.

3. Paragraph 2 sets forth the arbitral institution’s duty of oversight over third parties. Although artificial intelligence supporting arbitration may be independently developed and deployed by the arbitral institution, given the technical complexity involved, it is also possible in practice for the institution to collaborate with third parties that provide such services. Even if the arbitral institution does not itself serve as the direct provider of AI‑based services, it nonetheless bears an obligation to supervise the service provider.

The arbitral tribunal may appropriately employ artificial intelligence to assist in the arbitration proceedings, thereby enhancing procedural efficiency and the quality of the award.

1. This provision aims to clarify the legitimacy of the arbitral tribunal’s use of artificial intelligence.

2. The role of artificial intelligence in arbitration is that of a tool to enhance efficiency, rather than a decision‑making entity. Arbitral tribunals may leverage AI to handle auxiliary and administrative tasks—such as preliminary evidence organization, drafting correspondence, and supporting legal research—thereby enabling them to devote their specialized resources more effectively to core adjudicative functions, including hearings, deliberations, and legal analysis. For example, Article 7 of the South African Arbitrators’ Association’s “Guidelines on the Use of Artificial Intelligence in Arbitration and Awards” outlines numerous scenarios in which arbitral tribunals, parties, and counsel may employ AI during the arbitration process, including…

7.1 Analyze and integrate facts; 7.2 Conduct research and analysis; 7.3 Review and manage documents; 7.4 Accelerate the decision-making process; 7.5 Generate textual content; 7.6 Facilitate procedural workflows; 7.7 Automate administrative tasks.

The arbitral tribunal should exercise caution in the use of artificial intelligence:

(1) Understand the characteristics, limitations, and capability boundaries of artificial intelligence, and recognize that its outputs may contain inaccuracies or biases.

(2) Exercise due diligence in evaluating the outputs of artificial intelligence; do not treat them as the sole source of truth or infallible information. Instead, conduct independent assessments and verifications of their accuracy, completeness, reliability, and potential biases.

1. This provision aims to clarify that arbitral tribunals must exercise due diligence when employing artificial intelligence to assist in arbitration. This is because generative AI, built on large language models and trained on vast corpora, generates content by making probabilistic predictions based on input prompts. Such a mechanism inherently entails limitations, including the potential for “hallucinations,” biases, and information lags. In this regard, Clause 4 of the UK Courts and Tribunals Judiciary’s “Guidance on the Use of Artificial Intelligence for Judicial Personnel (2025)” advises: “AI tools based on large language models produce outputs that depend on their training datasets. The information generated by AI will inevitably reflect errors and biases present in those datasets, even if certain alignment strategies may help mitigate these issues. You should always be aware of this possibility and take steps to address it. (……)”

2. The “duty of prudence” set forth in this provision encompasses two distinct dimensions: first, cognitive prudence, requiring the arbitral tribunal to recognize that artificial intelligence is not an authoritative source of information and that its outputs are probabilistic constructs based on algorithms and training data, which may not be entirely accurate; second, behavioral prudence, mandating that the tribunal maintain a high degree of professional skepticism toward any materials generated by AI, refraining from undue credence or blind conformity, and conducting an independent assessment of the reliability of the resulting outputs.

The arbitral tribunal shall exercise due diligence and, notwithstanding the use of artificial intelligence, shall not neglect its adjudicatory duties; it must maintain independent analysis and judgment in ascertaining the facts of the case and applying the law.

1. This provision aims to clarify that, when employing artificial intelligence to assist in arbitration, the arbitral tribunal must exercise due diligence and may not, on the grounds of technological convenience, neglect its adjudicatory duties. As the decision-maker in a dispute, the tribunal must ensure that it devotes sufficient time and effort to the examination of the case and may not delegate its responsibilities entirely to AI. For example, Article 6 of the “Guidelines for the Use of Generative Artificial Intelligence by Judges, Judicial Officers, and Support Staff of the Judiciary of Hong Kong (2024)” stipulates: “Judges and judicial officers shall ensure that all judicial decisions continue to be rendered independently and personally by them, and under no circumstances shall generative artificial intelligence be permitted to substitute for or interfere with the performance of their judicial functions. In other words, the court must ensure that any use of generative artificial intelligence is purely supportive and intended to facilitate—rather than usurp or disrupt—their judicial mandate.”

2. The arbitral tribunal’s independent and professional judgment is the cornerstone of the arbitration system. With respect to core functions that involve the exercise of discretionary authority—such as fact-finding and the application of law—the tribunal must not rely exclusively on content generated by artificial intelligence. In this regard, Article 2.2 of the Vienna International Arbitral Centre’s “Guidelines on the Use of Artificial Intelligence in Arbitral Proceedings” expressly provides: “The use of artificial intelligence tools by arbitrators shall not substitute for their independent analysis of the relevant facts and legal issues; such tools may be used solely to assist in conducting that analysis, but not to replace it.”

The arbitral tribunal must review and verify content generated by artificial intelligence:

(1) With respect to any outputs generated by artificial intelligence—such as case summaries, legal research, or evidence compilations—these shall not be relied upon as the sole or decisive basis for analysis and judgment. Instead, they must be cross‑checked and verified by consulting the original materials, statutory texts, and other reliable sources.

(2) The output of one artificial intelligence system should not be used solely to verify the accuracy of the output of another artificial intelligence system.

1. This provision aims to establish the arbitral tribunal’s obligation to review and verify content generated by artificial intelligence.

2. The first provision establishes the principle of “no verification, no use.” Given the risk of unreliability inherent in AI-generated content, arbitrators should cross‑check and verify such content against primary, reliable sources, including case file evidence, the original texts of laws and regulations, and authoritative judicial precedents. AI‑generated analyses may not serve as the decisive basis for an award. For example, the American Arbitration Association–International Centre for Dispute Resolution states in the “Precautions for Using Artificial Intelligence” section of its “Guidelines for Arbitrators on the Use of Artificial Intelligence Tools”: “1. Accuracy and Reliability of Information: While AI tools can provide valuable assistance, they may sometimes produce incomplete or inaccurate information. Arbitrators must exercise their professional judgment to carefully evaluate and verify the outputs of AI tools, ensuring that the relevant information meets the standards of accuracy and reliability required by the arbitral procedure. When using AI tools, arbitrators should cross‑check their outputs against original documents or primary sources to conoffice the accuracy of the information.”

3. The second measure aims to prevent ineffective cyclical validation: one should not rely solely on an AI system that may produce “hallucinations” to verify the output of another similar tool. Such an approach cannot ensure the accuracy of the results and may even perpetuate errors. The ultimate benchmark for cross‑validation must be a widely recognized, reliable source of information. In this regard, Article 5(3)(e) of the Supreme Court of Singapore’s “Guidelines on the Use of Generative AI Tools by Courts (2024)” stipulates: “Please note that merely asking a generative AI tool to conoffice whether certain materials exist or whether they contain the content claimed to have been generated by AI does not constitute sufficient verification. Specifically, a generative AI tool must not be used to verify or conoffice the content generated by another generative AI tool.”

Arbitral tribunals are encouraged, prior to the conclusion of the first hearing, to disclose, in an appropriate manner, any matters relating to the use or planned use of artificial intelligence that may have a substantial impact on the conduct of the proceedings, and to afford the parties the opportunity to comment thereon. If a party expressly objects to the use of artificial intelligence, the arbitral tribunal shall take the party’s views into account and may issue corresponding procedural orders regarding such use.

1. This provision encourages arbitral tribunals to engage in disclosure and communication with the parties regarding the use of artificial intelligence, thereby safeguarding the parties’ right to information and enhancing the transparency of the arbitral process. For example, Clause 9.1 of the Chartered Institute of Arbitrators’ “Guidance on the Use of Artificial Intelligence in Arbitration (2025)” provides: “Unless otherwise agreed by the parties, arbitrators are encouraged to consult with the parties concerning the use of any artificial intelligence tool and to afford the parties an opportunity to comment on and object to the arbitrator’s use of such tool. If the parties disagree on the arbitrator’s use of an AI tool, the arbitrator should refrain from using that particular AI tool. If the parties consent to the arbitrator’s specific use of an AI tool, the arbitrator is encouraged to make a decision on whether to proceed, taking into account all relevant circumstances of the case, including the considerations set out in paragraphs 8.1 through 8.4 above.”

2. This provision clarifies the threshold for disclosure, namely “that which may have a material impact.” For example, using artificial intelligence to assist in drafting pleadings or performing spell‑checking generally does not require disclosure; however, if the plan is to employ AI to analyze large volumes of evidence in order to help identify the key issues in dispute, such use must be disclosed.

3. This provision specifies that the arbitral tribunal must, at the latest, disclose information and engage in communication with the parties before the conclusion of the first hearing; this does not preclude the tribunal from conducting such disclosure and communication prior to that point. The provision further clarifies that the methods of disclosure and communication are to be “appropriate,” encompassing both oral and written forms.

4. This article establishes a “disclosure–communication–decision” procedure. After hearing the parties’ submissions, the arbitral tribunal is empowered to render a final award and may, by way of a procedural order, specify the relevant arrangements, such as the tools permitted for use and the methods of data processing.

The secretary of the arbitral tribunal may appropriately employ artificial intelligence in assisting the tribunal’s work, subject to the same standards as those applicable to the tribunal itself.

1. This provision aims to clarify that the use of artificial intelligence by the arbitral tribunal’s secretary shall be governed by the same standards applicable to the tribunal itself. As the secretary’s role is to assist in advancing the arbitral proceedings under the tribunal’s direction and supervision, any use of AI by the secretary may be regarded as an extension of the tribunal’s actions and must accordingly adhere to requirements such as prudent application, diligent performance of duties, and thorough review and verification. For example, Article 10 of the “Arbitral Secretary Standards (Recommended Text)” jointly drafted by the Institute of International Law of the Chinese Academy of Social Sciences and the Guangzhou Arbitration Commission stipulates: “The arbitral tribunal’s secretary shall observe the same standards of impartiality and independence as the arbitrators.” Similarly, with respect to the regulation of AI use, the tribunal’s secretary should maintain consistency with the tribunal.

The parties may agree on the manner, scope, and limitations of the use of artificial intelligence in arbitration, provided that such agreements do not contravene applicable laws and regulations.

1. This provision expressly stipulates that the parties may freely agree on matters relating to the use of artificial intelligence in arbitration, thereby embodying the principle of party autonomy in arbitral proceedings; however, such agreements must not contravene mandatory legal provisions. Similarly, Article 5.1 of the Chartered Institute of Arbitrators’ “Guidance on the Use of Artificial Intelligence in Arbitration (2025)” adopts a comparable stance, providing: “Subject to compliance with applicable laws, mandatory rules, and the terms of the arbitration agreement, the parties may independently determine whether to employ artificial intelligence, which specific AI tools to use, and any applicable limitations or conditions.”

2. The parties’ agreed-upon provisions may cover various aspects, including the manner of use, scope, and limitations: the manner of use primarily pertains to the specific application scenarios and operational procedures of artificial intelligence; the scope of use may delineate the stages, procedural steps, and matters in which AI is engaged; and the limitations on use chiefly specify prohibitive or restrictive conditions governing the deployment of AI technologies and tools.

The parties and their representatives shall be familiar with the basic functions and potential risks of the artificial intelligence systems they employ, review any content generated by such systems in the course of arbitration, and assume responsibility for the evidentiary value of that content.

1. This provision clarifies the due diligence and review obligations of the parties and their agents when using artificial intelligence.

2. With respect to the duty of due care, the parties and their representatives shall familiarize themselves with the basic functions of the artificial intelligence system being used—including its technical principles, applicable scenarios, and output formats—as well as its potential risks—such as possible errors, biases, and limitations—and, on that basis, exercise prudent use to prevent any improper application from infringing upon their own rights and interests or disrupting the proper conduct of the arbitral proceedings.

3. With respect to the duty of review, the parties and their legal representatives are required to examine and verify the authenticity, accuracy, and appropriateness of AI-generated content used in arbitration. If the parties or their representatives fail to fulfill their duties of due diligence and review, they shall bear the adverse consequences of the evidence being deemed inadmissible. For example, the Canadian Federal Court’s “The Use of Artificial Intelligence in Court Proceedings (2024)” explicitly states in its section on “Principles for the Use of Artificial Intelligence” that “it is essential to scrutinize documents and materials generated by AI, and the court urges that all AI‑generated content be subject to verification.”

The parties and their agents shall not use artificial intelligence to fabricate or tamper with evidence.

1. This provision establishes a prohibition on the use of artificial intelligence by parties and their agents.

2. Artificial intelligence possesses the technical capabilities to generate, edit, and modify content, which may be misused to fabricate false evidence or tamper with authentic evidence. This provision seeks to prevent such conduct through prohibitive measures, thereby safeguarding the authenticity of evidence. The prohibited AI‑related activities encompass both fabrication and alteration. Fabrication of evidence refers to the use of AI to create entirely fictitious evidentiary materials, including the generation of false documents, data, images, audio, video, and other forms of evidence. Alteration of evidence involves employing AI to make substantial modifications to existing, genuine evidentiary materials, thereby changing their original state or true content and rendering them inconsistent with objective facts. For example, Guideline 5 of the Silicon Valley Center for Arbitration and Mediation’s “Guidelines on the Use of Artificial Intelligence in Arbitration” explicitly states: “Parties, counsel, and experts shall not, in any form, use artificial intelligence to fabricate evidence or undermine the authenticity of evidence.”

3. The entities obligated under this provision are the parties to the proceedings and their legal representatives. The parties shall bear responsibility for the authenticity of the evidence they submit, and the legal representatives likewise have an obligation to ensure the authenticity of such evidence.

When a party submits evidence generated by artificial intelligence, it shall clearly label such evidence and notify the arbitral tribunal, and provide usage records that enable an assessment of its probative value, so that the tribunal may render a determination regarding that evidence.

1. This provision sets forth the obligation of parties to clearly identify evidence generated by artificial intelligence.

2. At present, there already exist evidence types generated or processed by artificial intelligence, such as risk assessment reports and statistical analysis conclusions produced with AI, operational logs of AI systems, and electronic forensic evidence derived from AI technologies. These forms of evidence differ from traditional evidence in terms of their mode of creation and reliability. Accordingly, this provision requires the parties to fulfill an obligation to label such evidence and to provide usage records that enable an assessment of its probative value, thereby facilitating the arbitral tribunal’s identification of these evidentiary materials and the adoption of appropriate review procedures for accurately evaluating their evidentiary weight. For example, Article 5(4) of the Supreme Court of Singapore’s “Guidelines on the Use of User-Generated Artificial Intelligence Tools in Courts (2024)” explicitly stipulates: “You must be able to identify which specific portions of court documents contain content generated by artificial intelligence and explain to the court how you have verified the output of the generative AI tool.” Furthermore, Article 10 of the “Methods for Labeling AI-Generated Content in Cybersecurity Technology,” jointly issued by China’s Cyberspace Administration of China, the Ministry of Industry and Information Technology, and other authorities, mandates: “Users who publish generatively synthesized content through online information dissemination services shall proactively declare such content and apply the labeling function provided by the service provider. No organization or individual may maliciously delete, alter, falsify, or conceal the labels for generatively synthesized content prescribed in these Measures; nor may they provide tools or services to others for engaging in such malicious acts, or infringe upon the legitimate rights and interests of others through improper labeling practices.”

3. The identification obligations set forth in this Article comprise three elements. First, parties must affix conspicuous markings, requiring them to clearly indicate on the evidentiary materials that the evidence was generated or processed by artificial intelligence; such markings must be readily perceptible. Second, parties are required to inform the arbitral tribunal, proactively disclosing to it the manner in which the evidence was created. Third, parties must provide usage records, furnishing relevant information necessary to assess the reliability of the evidence, including the name and version of the AI system employed, the time of use, the input data, and the output results. Evaluating the probative value of such evidence necessitates understanding the AI’s application context as well as any potential errors or biases; accordingly, the usage records submitted by the parties must be complete and detailed, sufficient to enable the arbitral tribunal to render a reasoned determination regarding the evidence’s probative force.

Parties and their counsel are encouraged, prior to the conclusion of the first hearing, to proactively disclose to all parties and the arbitral tribunal any matters relating to the use or planned use of artificial intelligence that may have a substantial impact on the proceedings.

1. This provision establishes an incentive‑based disclosure regime for the use of artificial intelligence, with the aim of enhancing transparency in arbitral proceedings. Such disclosure enables the arbitral tribunal and all parties to promptly ascertain the extent of AI utilization, thereby allowing them to anticipate and address potential risks. The preface to the Canadian Federal Court’s “The Use of Artificial Intelligence in Court Proceedings (2024)” states: “The court expects that, when submitting pleadings to the court, any party shall notify the court and the other parties if those submissions contain content generated by artificial intelligence.”

2. The scope of disclosure is determined primarily by the criterion of “substantial impact,” and not all uses of artificial intelligence are subject to disclosure. In practice, disclosure may be required when the application of AI could significantly affect the assessment of evidence or the outcome of a decision, or in key situations involving the explicit delegation of duties to arbitrators and other relevant parties. Article 7.1 of the Chartered Institute of Arbitrators’ “Guidance on the Use of Artificial Intelligence in Arbitration (2025)” likewise stipulates that disclosure is necessary only when the use of an AI tool may influence the evidence or the arbitral outcome. This limitation prevents an unwarranted expansion of the disclosure obligation, striking a balance between procedural efficiency and informational symmetry.

3. The timing of disclosure is set to before the conclusion of the first hearing, balancing procedural efficiency with the protection of rights. Completing disclosure prior to the first hearing ensures that all parties are fully informed and can promptly submit their views before the substantive examination of the case, while also enabling the arbitral tribunal to adjust its hearing plan and procedural arrangements in a timely manner, thereby avoiding procedural disputes, delays in proceedings, or the risk of surprise arising from disclosure during or after the proceedings.

Where the use of artificial intelligence may result in substantive harm to the legitimate rights and interests of the parties, the parties shall have the right to submit a written request to the arbitral tribunal or the arbitration institution for review. The arbitral tribunal or the arbitration institution may decide to adopt any necessary measures.

1. This provision sets forth the remedies available to parties in cases of improper use of artificial intelligence and clarifies their corresponding rights of claim.

2. The prerequisite for a party to seek relief is that the use of artificial intelligence has the potential to cause substantial harm to its legitimate rights and interests, either in the arbitral proceedings or in the outcome of the award. Here, “substantial harm” means harm of such a degree as to affect the party’s fundamental rights or the correctness of the award, excluding minor or technical defects. By adopting a precautionary standard based on the possibility of harm, rather than requiring that harm have already occurred, this approach facilitates the timely implementation of measures to prevent further damage. For example, Clause 4.7 of the Chartered Institute of Arbitrators’ Guidelines on the Use of Artificial Intelligence in Arbitration (2025) provides: “If a party fails to comply with instructions or procedural orders regarding the use of artificial intelligence, the arbitrator shall assess the impact of such non‑compliance on the arbitral proceedings. Thereafter, the arbitrator may take any measures to remedy the non‑compliance, make any further rulings concerning the use of artificial intelligence, draw any appropriate conclusions—including, where appropriate, a negative presumption—or take the non‑compliance into account when determining costs.” Similarly, Clause 7.7 stipulates: “The arbitrator shall assess any impact of undisclosed conduct on the fairness of the arbitral proceedings. Thereafter, the arbitrator may take any measures to address such undisclosed conduct, make any further rulings concerning the use of artificial intelligence, draw any appropriate conclusions—including, where appropriate, a negative presumption—or consider the undisclosed conduct when determining costs.”

3. The parties have the right to submit a request for relief in writing to the arbitral tribunal or the arbitral institution. The application shall clearly set forth the circumstances of the use of artificial intelligence, the issues at stake, the harm that has been caused or may be caused, and the specific measures sought. A written form helps ensure the clarity and completeness of the request, facilitating review and decision‑making by the arbitral tribunal or the arbitral institution. In institutional arbitration, the parties may choose to address their request either to the arbitral tribunal or to the arbitral institution; in ad hoc arbitration, the parties may only submit their request to the arbitral tribunal.

4. The arbitral tribunal or arbitration institution accepting the application shall assess the extent to which the use of artificial intelligence affects the arbitral proceedings or the outcome of the award, and decide whether to adopt appropriate measures. In exercising its discretion, the tribunal or institution shall take into account such factors as the degree of impropriety in the use of artificial intelligence, the scope of its impact on the adjudication of the case, the fault attributable to the parties, and the feasibility of remedial measures, and render a reasoned decision that safeguards procedural fairness and the accuracy of the award.

Unless the parties have otherwise agreed or the arbitration rules provide otherwise, this Guideline is not legally binding, does not form part of the arbitration rules, does not affect the validity of an arbitral award, and does not constitute a basis for judicial review of the arbitration.

In the event of any inconsistency between this Guideline and the Arbitration Rules, the Arbitration Rules shall prevail.

This Specification Guide is issued and interpreted by XXX.

This Standard was issued on X month X day, X year, and shall take effect from the date of its issuance.

Research Group Information

Principal Investigator:

Mao Xiaofei, Deputy Director of the Private International Law Research Office at the Institute of International Law, Chinese Academy of Social Sciences; Associate Researcher; and Member of the Ministry of Justice’s Expert Committee on Foreign-related Rule of Law.

Li Minshen, Deputy Director of the Guangzhou Arbitration Commission

Lu Bin, General Counsel and Chief Compliance Officer of Inspur Digital Enterprise Technology Co., Ltd.

Zhou Xin, former Deputy General Manager and Chief Legal Counsel of Beijing Huayu Yuandian Information Service Co., Ltd.

Zhu Yunyang, Director and Managing Partner of Beijing Qinglv Law Office

Project team members:

Xu Bing, Comprehensive Secretary of the Research Office, Guangzhou Arbitration Commission

Du Keyu, Legal and Compliance Manager, Inspur Digital Enterprise Technology Co., Ltd.

Liu Guiying, Legal and Compliance Manager, Inspur Digital Enterprise Technology Co., Ltd.

Zhang Meng, Legal and Compliance Manager, Inspur Digital Enterprise Technology Co., Ltd.

Xu Ke, formerly a Senior Knowledge Engineer at Beijing Huayu Yuandian Information Service Co., Ltd.

Zhou Ying, Senior Partner at Beijing Qinglv Law Office

Qiu Shanshan, Research Assistant at Beijing Qinglv Law Office

Wang Fenghe, Senior Partner at Beijing Yingke Law Office and Director of the Yingke Beijing Digital Economy Law Research Association.

Liu Rundong, Partner at Deheheng Law Office in Beijing

Project Advisory Committee:

Liu Huawen, Director and Researcher at the Institute of International Law, Chinese Academy of Social Sciences

Wang Tianxi, Secretary of the Party Committee and Director of the Guangzhou Arbitration Commission

Research Assistant:

Lü Silei, graduate student at the Graduate School of the University of the Chinese Academy of Social Sciences.

Shi Haixiao, a graduate student at the Graduate School of the University of the Chinese Academy of Social Sciences.