DWAC Feedback on China's AI Agent Governance Framework
Executive Summary
DWAC submits feedback on China's Implementation Opinion on AI Agent Standardization and Innovation Development (May 8, 2026), proposing four key recommendations: (1) refining "clear decision boundaries" principles, (2) improving the security assessment framework, (3) supplementing multi-agent and cross-border interoperability rules, and (4) establishing pre-arbitration dispute resolution mechanisms.
About DWAC
The Digital World Arbitration Center (DWAC) is an online arbitration institution specializing in digital world dispute resolution, headquartered in the Hong Kong Special Administrative Region of the People's Republic of China. DWAC is committed to providing efficient and impartial arbitration services for cross-border digital disputes, having established an arbitration rules system covering 47 jurisdictions worldwide and accumulated extensive institutional research in artificial intelligence arbitration.
Overall Assessment
DWAC warmly welcomes the release of this Implementation Opinion and recognizes its threefold significance:
- ●Filling an international gap in Agent-specific governance: China is the first major jurisdiction globally to issue a systematic governance document specifically targeting AI Agents, demonstrating distinct institutional foresight.
- ●Establishing an effective paradigm for multi-departmental collaborative regulation: Jointly issued by CAC, NDRC, and MIIT, reflecting institutional innovation in cross-departmental AI governance coordination.
- ●Clarifying core principles of security and controllability: "Security and controllability" as the primary principle for AI Agent governance aligns with the international consensus on high-risk AI system regulation.
Key Recommendations
1. Refining "Clear Decision Boundaries" Principles
The Implementation Opinion requires "clearly delineated decision-making authority boundaries" between users and AI Agents.
Recommendations:
- • Establish classification and grading standards for AI Agent decision-making authority (e.g., high-risk decisions requiring human confirmation)
- • Require AI Agent operators to disclose the scope of autonomous decision-making and authorization boundaries to users
- • Establish an "Agent operator vicarious liability" regime, clarifying operators' legal responsibility for Agent behavior
2. Improving Security Assessment Framework
The Implementation Opinion requires AI Agent operators to submit "security assessment reports" before deployment.
Recommendations:
- • Issue "AI Agent Security Assessment Technical Specifications" clarifying assessment indicator systems, processes, and qualification standards
- • Establish a recognition system for "independent third-party assessment institutions"
- • Clarify in arbitration rules that security assessment reports can serve as important evidence for determining whether operators have fulfilled reasonable duty of care
3. Multi-Agent Systems & Cross-Border Interoperability
The Implementation Opinion envisions multi-agent systems and proposes building a "trusted interconnection foundation."
Recommendations:
- • Clarify joint liability frameworks and operator identification obligations in multi-agent collaboration scenarios
- • Establish jurisdiction rules for cross-border AI Agent disputes, with "service results directed at Chinese users" as a jurisdictional connecting factor
- • Promote interoperability between AID identity codes and W3C Verifiable Credentials and international A2A/MCP protocol standards
4. Pre-Arbitration Dispute Resolution Mechanism
AI Agent disputes are highly technical, professional, and cross-border in nature, with traditional litigation facing challenges of evidentiary difficulty and high cross-border enforcement costs.
Recommendations:
- • Add encouraging provisions in implementation rules, guiding AI Agent operators to include arbitration clauses in service agreements
- • Require AI Agent operators to preserve complete decision trail logs as key evidence in arbitration proceedings
- • Establish "AI Arbitrator" certification system, allowing qualified AI systems to participate in dispute resolution as technical assistance tools under specific conditions
Conclusion
DWAC believes this Implementation Opinion marks a new stage in China's AI governance development. The feedback provided aims to contribute to the refinement of this important policy framework. DWAC looks forward to maintaining communication with CAC and relevant departments, providing professional support in the field of digital arbitration.
Contact: Harry Zhang, Founder
Email: contact@dwac.net
Date: September 1, 2026
This feedback was prepared by DWAC's technical team based on discussions among Dr. Frontie, Pr. Tc Zhou, Fresa Li, and Dr. Gochye.
Related Publications
Explore DWAC's research on AI governance and digital arbitration.
View Global Cyber Law Compendium