Editor’s Note — DCC.
What this is. A voluntary initiative (倡议), not a law, a standard or a regulator’s document. It binds no one who has not joined it, and says so. DCC translates it because it is the first multi-institution statement in China of what the rules between an AI agent, the user who authorizes it and the app it operates ought to be, and because it appeared the day after MIIT’s Agent Terminal Notice (工信厅信管〔2026〕42号) settled one of those questions by rule.
Where it comes from. It is the “consensus declaration” Li Wenlong announced at the end of his essay on the SAEP refusals, translated in The Agent Authorization Standoff After SAEP.
The source and the translation. The WeChat post carries the Initiative as nine page images with no text layer. DCC read them by OCR and checked the result against the images. The ten principles and the ten questions and answers are translated in full, as are the front-matter sections. The list of drafting experts is given by name and affiliation; honorifics and secondary titles are shortened. The contact address for joining is in the original and is not reproduced here. English names of institutions are DCC’s renderings unless the institution uses one of its own.
Initiative on Basic Rules for Agent Authorization and Interoperability
智能体授权互操作基础规则倡议 · Version V1.0
Front matter
I. Purpose of the Initiative
This Initiative aims to promote trusted interoperability of agents across applications, platforms and ecosystems and, on the basis of respecting users’ autonomous choice, safeguarding lawful rights and interests and maintaining security and fair competition, to advance the building of open, non-discriminatory and transparent interoperability rules and mechanisms for compatibility and mutual recognition.
II. Who it is addressed to
This Initiative is addressed to those who take part in agent interoperability and in the related rule-making, technical implementation and governance work, including but not limited to:
- Agent service providers: those who develop, deploy or operate agents or provide related agency services.
- Application and service operators: those who provide applications, platforms, digital services and related resources that agents can access, invoke or operate.
- Terminal and system providers: those who provide smart terminals, operating systems, runtime environments, identity authentication, permission management or policy enforcement capabilities.
- Technical protocol and infrastructure providers: those who provide interoperability protocols, interface standards, authentication, tool calling and other technical support.
- Other relevant parties: including industry associations, standardization organizations, research institutions, representatives of users and of consumer interests, and other organizations or individuals taking part in research, standard-setting, technical evaluation, industry collaboration and the promotion of practice in agent interoperability.
III. Scenarios covered
This Initiative applies mainly to conduct in which an agent, on the basis of the user’s valid authorization, accesses digital resources, processes information and performs tasks across applications, platforms or services by screen interaction, application interfaces, tool calling or other technical means. Specifically, it covers agent identity recognition, authorization and permission management, the expression of access conditions, policy negotiation and enforcement, risk control, operation records, verification of results, dispute handling and cooperation on liability.
IV. How to take part
This Initiative is published openly and participation is voluntary. Those wishing to take part may submit their intention through the participation registration channel. [A contact address follows in the original.]
V. Release and effect
This Initiative was formally released on November 7, 2026 at the Sixth Forum on Rule of Law and Reform (第六届法治与改革论坛) held at Zhejiang University, in the sub-forum “Legislative Choices and Rule Design for Agents” (智能体的立法选择与规则设计).
VI. Revision
The initiating unit is responsible for organizing day-to-day communication, the collection of opinions, coordination of implementation and version maintenance for this Initiative. Later revisions shall adhere to the principles of openness, transparency and multi-party participation, and shall fully hear the views of agent providers, application operators, terminal and system providers, user representatives and other stakeholders. Major revisions shall in principle be accompanied by a reasonable consultation and transition arrangement.
VII. Participating bodies
Initiating unit: Zhejiang University Institute of Digital Rule of Law (浙江大学数字法治研究院).
Co-initiating units, as listed in the original:
- Research Center for Digital Economy and Legal Innovation, University of International Business and Economics (对外经济贸易大学数字经济与法律创新研究中心)
- Shanghai Collaborative Innovation Center for AI Social Governance, Tongji University (同济大学上海市人工智能社会治理协同创新中心)
- Global AI Innovative Governance Center (全球人工智能创新治理中心)
- Research Center for AI and Information Security Law, Xi’an Jiaotong University (西安交通大学人工智能与信息安全法律研究中心)
- MIIT Key Laboratory of Smart Rule of Law for Industry and Information Technology (工业和信息化智慧法治工信部重点实验室)
- Zhejiang University Institute of Digital Rule of Law (listed again among the co-initiators)
- Hainan University Law School (海南大学法学院)
- Institute of AI Law, China University of Political Science and Law (中国政法大学人工智能法研究院)
- Data Law Research Center, East China University of Political Science and Law (华东政法大学数据法律研究中心)
- King & Wood Mallesons, Beijing (北京市金杜律师事务所)
- Jinan University Law School (暨南大学法学院)
- Shihui Partners, Beijing (北京世辉律师事务所)
- Research Center for Competition Law and Intellectual Property, Shanghai University of Finance and Economics (上海财经大学竞争法与知识产权研究中心)
- Tahota Law Firm (泰和泰律师事务所)
- Center for AI Governance and Law, Shanghai Jiao Tong University (上海交通大学人工智能治理与法律研究中心)
- Zhejiang Kending Law Firm (浙江垦丁律师事务所)
- Institute of Law and Technology, Shenzhen University (深圳大学法律与科技研究院)
Drafting experts (26, in the original’s alphabetical order by pinyin):
| Name | Affiliation as listed |
|---|---|
| Cao Jianfeng (曹建峰) | Shenzhen University Law School |
| Chen Tianhao (陈天昊) | Tsinghua University School of Public Policy and Management; Tsinghua Center for Science and Technology Development and Governance |
| Feng Yang (冯洋) | Zhejiang University Guanghua Law School; ZJU Institute of Public Law and Comparative Law |
| Gao Fuping (高富平) | East China University of Political Science and Law, data rule-of-law research center |
| Li Wenlong (李汶龙) | Zhejiang University Guanghua Law School; member of a UNESCO AI ethics committee |
| Li Zhenhua (李振华) | Ant Group Research Institute |
| Liao Huaixue (廖怀学) | Tahota Law Firm; Renmin University Law School (doctoral candidate) |
| Liu Xiaochun (刘晓春) | University of Chinese Academy of Social Sciences, Internet Rule of Law Research Center |
| Ma Ce (麻策) | Kending Law Firm |
| Ma Changshan (马长山) | East China University of Political Science and Law, Institute of Digital Rule of Law |
| Pan Ning (潘宁) | Jinan University Law School |
| Wang Xinlei (王新雷) | Xi’an Jiaotong University Law School; Research Center for AI and Information Security Law |
| Wang Xinrui (王新锐) | Shihui Partners |
| Wu Han (吴涵) | King & Wood Mallesons |
| Wu Peicheng (吴佩乘) | Zhejiang University Guanghua Law School |
| Xu Ke (许可) | University of International Business and Economics Law School |
| Yang Fang (杨芳) | Hainan University Law School |
| Yao Xu (姚旭) | Global AI Innovative Governance Center; Fudan Development Institute |
| Yuan Bo (袁波) | Shanghai University of Finance and Economics |
| Zhang Jiyu (张吉豫) | Renmin University Law School; Institute for Future Rule of Law |
| Zhang Linghan (张凌寒) | China University of Political Science and Law, Institute of AI Law |
| Zhang Weijun (张伟君) | Zhejiang University Guanghua Law School |
| Zhang Xin (张欣) | University of International Business and Economics Law School |
| Zhao Jingwu (赵精武) | MIIT Key Laboratory of Smart Rule of Law for Industry and Information Technology |
| Zhao Zerui (赵泽睿) | Shanghai Jiao Tong University, Center for AI Governance and Law |
| Zhu Lingfeng (朱玲凤) | Legal expert at an internet company |
Acknowledgments. From drafting to final text, this Initiative went through many rounds of discussion and revision, and benefited from the wisdom and support of many people from industry, academia, legal practice, technical standards and related fields of public policy research. Special thanks go to Hong Yanqing (洪延青), Zhu Yue (朱悦), Xu Peng (徐鹏), Hu Naying (呼娜英), Xu Gang (徐钢), Fu Hongyu (傅宏宇), Kang Weibing (康维兵), Wu Shaoqing (吴少卿) and other experts and scholars (in no particular order) for their professional support and careful advice. In addition, some experts, industry colleagues and other participants were unable or found it inconvenient to sign publicly, by reason of professional duties, institutional requirements, work arrangements or otherwise, but made important contributions to the conception of the rules, the technical argument and the coordination of multiple interests. We fully respect their wishes as to attribution and the related confidentiality requirements, and extend sincere thanks for the trust, support and professional commitment they have given. We look forward to continuing, on the basis of mutual respect and open collaboration, to bring together the experience of all parties and jointly to advance the continuing improvement of the rules for agent authorization and interoperability.
Version number: V1.0
The ten principles
1. Autonomous authorization (自主授权)
An agent accesses third-party services within the scope of its lawful permissions, on the basis of the user’s genuine, explicit and valid authorization or of another basis expressly provided by law. The agent’s agency permissions shall be commensurate with the rights the user enjoys by law and the scope of what the user may authorize, and are subject to the provisions of law on personal information protection, cybersecurity, transaction security and other matters.
An agent’s acts of agency shall make clear the authorizing party, the matter, the scope of permission and the duration, and shall support the user in querying, adjusting, suspending and withdrawing them. For operations involving identity authentication, control of physical devices, major disposals of property, personal safety and the like, mechanisms for separate confirmation, user takeover or limits on automatic execution shall be set according to risk.
2. Proportionate interoperability (适度互操作)
It is advocated that a proportionate balance be sought among users’ reasonable use, the opening of services and security safeguards: providing agents with the access and operating capability necessary to complete authorized tasks, while fully respecting the security, resources and reasonable business needs of third-party services.
All parties are encouraged to set the scope of access, execution permissions and conditions for continuous operation reasonably, according to the nature of the task, the degree of interoperability and the actual risk; to give priority to exploring flexible measures such as graded authorization, capability limits and user confirmation; and to reduce unnecessary obstacles to access while safeguarding security.
3. Transparent rules (规则透明)
It is advocated that the access conditions, restrictive measures and related rules of agent interoperability be open, transparent, clear and easy to understand, so that all participants can reasonably anticipate the conditions of access and collaboration.
The relevant parties are encouraged to provide appropriate explanations and complaint channels for important restrictions, and to communicate major rule changes in advance, leaving all parties a reasonable time to adjust and adapt.
4. Open standards (开放标准)
It is advocated that, on the basis of existing open protocols and technical standards, standardization and compatible mutual recognition be advanced in key links such as agent identity recognition, authorization credentials, capability description and task interaction, so as to promote coordinated operation among different agents, terminals and application services.
The adoption of open and extensible technical specifications is encouraged, supporting connection between different technical paths and equivalent implementations, and lowering the adaptation cost of cross-platform and cross-ecosystem collaboration.
5. Reasonable non-discrimination (合理非歧视)
All parties shall follow the principle of reasonable non-discrimination and promote fair access by agents to the relevant services in accordance with laws and regulations. Third-party service operators may set reasonable conditions such as identity authentication, permission management, call frequency and fee arrangements, on the basis of legitimate needs such as account security, legal compliance, system capacity, reasonable cost and the order of transactions. The conditions shall have an objective basis and be commensurate with the actual risk and the resources consumed.
6. Equal treatment (平等待遇)
In making and implementing agent interoperability rules, reasonable commercial arrangements and security-management needs shall be respected, and fair and foreseeable conditions of cooperation created for those taking part in cross-platform and cross-service collaboration.
Parties that hold the relevant access or rule-enforcement capability are encouraged to offer fair and reasonable opportunities for cooperation in comparable scenarios and under comparable risk conditions, to reduce unnecessary technical and ecosystem barriers, and to promote innovation and healthy competition.
7. Fair competition (公平竞争)
It is advocated that an open and fair environment for agent interoperability be fostered, that all parties’ reasonable commercial arrangements and security-management needs be respected, and that different players take part fairly and compete healthily in cross-platform and cross-ecosystem collaboration.
Parties that hold key interfaces, terminal entry points or ecosystem resources are encouraged to set access conditions reasonably, to take account both of their own business interests and of other participants’ opportunities to innovate, and to reduce unnecessary barriers to competition by means such as transparent negotiation and reasonable restriction.
8. Functional reciprocity (功能互惠)
It is advocated that agent providers, terminal and system providers and third-party service operators each bring their strengths to bear and, on the basis of voluntary negotiation, form reciprocal cooperation mechanisms commensurate with technical capability, resources invested and risk borne.
The parties promote interoperability through capability sharing, technical collaboration and shared responsibility. Reciprocity lies in a reasonable balance of contribution and responsibility in cooperation, not in requiring the parties to open the same resources or provide equivalent services.
9. Multi-party co-governance (多方共治)
Agent providers, third-party service operators, terminal manufacturers, standards organizations, user representatives and professional institutions are encouraged to take part together in making and improving rules on agent access, the enforcement of authorization, risk management and the coordination of disputes, and guidance from the competent authorities is welcomed. It is advocated that rule-making, enforcement and evaluation be coordinated with one another and appropriately separated in function.
Connection with domestic and international standards organizations, industry collaboration mechanisms and related governance dialogues is encouraged, to promote the exchange of practical experience and reduce unnecessary fragmentation of ecosystems.
10. Effective remedy (有效救济)
It is advocated that convenient and transparent mechanisms for handling interoperability disputes be established, giving users and the relevant participants channels for feedback, verification and correction on problems such as access restrictions and abnormal execution.
The relevant participants should make clear how matters are received and answered, and cooperate promptly to verify key facts and correct problems that have been found. For major or continuing disputes, the introduction of an independent third party to carry out technical evaluation and coordination may be explored.
Questions and answers on the core content of the Initiative
Question 1. After a user authorizes an agent, must a third-party service allow access?
Not as a matter of course. The access conditions for reasonable use of a service already obtained, for continuous execution across pages, and for commercial aggregation across services are different. Whether there is a specific obligation to grant access is to be judged by law, by valid contracts, and by the commitments undertaken on expressly joining this Initiative.
Question 2. If a third-party service has not declared a refusal, does that amount to permission?
No. A default of no restriction at the protocol level does not as a matter of course constitute consent or permission in law, or authorization of a specific transaction. A third-party service’s silence, lack of adaptation or failure to object is not the same concept as legal consent, permission for access, or authorization for a specific use of data.
Question 3. May an operator restrict agents on commercial grounds?
Reasonable service cost, carrying capacity and the order of transactions may be taken into account. But competition over the entry point, or reduced advertising exposure, is not by itself enough to support a restriction, and contracts, competition law and commitments already made on fair access must still be observed.
Question 4. Must all agents receive exactly the same treatment?
No. Agents that are similar in scope of authorization, technical capability, risk and trust conditions should receive fair treatment; real differences can support graded requirements that have a basis.
Question 5. How can a continuously running agent be kept within its authorization at all times?
A continuous task may run on explicit authorization. But where the object of the task, the amount, the use of data or similar matters change materially, or the operation enters a higher risk level, there should be fresh verification, confirmation or an offer of user takeover according to risk, and repeated execution should be guarded against. A withdrawal of authorization should also be passed on to downstream tasks not yet executed. For a high-risk operation whose result is unclear, the status should be checked first, to avoid executing it again.
Question 6. Does reciprocity mean that every party must open the same resources?
It does not. Functional reciprocity stresses that rights and responsibilities, investment and risk should be commensurate. It does not require the same resources to be opened, and it may not be used to exclude new entrants.
Question 7. How is one to judge whether a restriction is reasonable?
By an overall assessment of the legitimate purpose, the evidence of risk, the effectiveness of the measure, less intrusive alternatives, the effect on users and the market, and the conditions for lifting it.
Question 8. Is identical rule text enough to prove non-discrimination?
It is not. One must also compare identity authentication, the opening of capabilities, the actual path of policy enforcement, service quality and the outcome of remedies.
Question 9. How are disputes to be resolved and liability for harm determined?
The relevant parties first verify the authorization, the policy and the execution records under an open procedure, bringing in independent review where necessary. Liability for harm is determined according to law and contract, in light of the specific conduct, system defects, capacity for control, fault and causation. Participants should agree in advance on a collaborative process for receiving incidents, stopping loss, preserving evidence, verifying results and sharing reasonable costs.
Question 10. Can this Initiative directly require a company to open its interfaces?
It cannot. This Initiative can create contractual obligations through voluntary accession. Mandatory opening by operation of law requires a separate, express basis and the satisfaction of the corresponding statutory conditions.
DCC’s reading
The notes below are DCC’s, not the Initiative’s.
- On silence, the Initiative and MIIT now agree. Question 2 says a service’s failure to declare a refusal is not consent. That is a rejection of the opt-out premise of Doubao’s SAEP. Part I, item (4) of 工信厅信管〔2026〕42号 reaches the same result by rule: no declared consent, no forcible invocation.
- On reasons, the Initiative goes where the Notice does not. The Notice asks only whether the app developer has declared consent. Principles 5 to 7 and Questions 3, 7 and 8 ask whether a refusal or a condition has an objective basis, and say that lost ad exposure alone is not one. None of this binds an app developer that has not joined.
- It is drafted as soft law on purpose. Almost every operative sentence is “advocated” (倡导) or “encouraged” (鼓励). The main exceptions, where the text says “shall” (应), are Principle 1 on the content and revocability of the user’s authorization, Principle 5 on the objective basis for conditions, and the opening sentence of Principle 6. Question 10 states the limit: obligations arise by joining, as contract.
- Who has signed, and who has not. The initiators are university research centers and law firms. Two drafting experts are listed from industry: one from Ant Group’s research institute, and one described as a legal expert at an internet company. The list published with V1.0 names no handset maker, app platform or agent provider as an initiating unit. Whether any of them join will decide whether the Initiative matters in practice.
- For users’ personal information, Principle 1 subjects the agent’s permissions to personal information protection law. For how consent under the PIPL strains when an agent reads the screen, see AI Agents and the Limits of Consent.
— 智能体授权互操作基础规则倡议 (Initiative on Basic Rules for Agent Authorization and Interoperability), V1.0, released November 7, 2026; published by 科技利维坦 on November 9, 2026. Initiating unit: Zhejiang University Institute of Digital Rule of Law. Translated by DCC from the page images in the original post. Original (Chinese).
Names and affiliations are as listed in the original; romanizations and English institution names are DCC’s and may differ from those the persons and institutions use themselves.
— Not legal advice.