The rapid spread of AI agents, exemplified by OpenClaw’s swift market penetration, has accelerated China’s discourse on AI safety governance. Initially seen as innovative digital assistants, these agents revealed security vulnerabilities and ethical risks, drawing urgent attention from government agencies. On February 5, 2026, the Ministry of Industry and Information Technology’s (MIIT) National Vulnerability Database (NVDB) flagged certain OpenClaw versions as having “relatively high security risks,” prompting government advisories. By mid-March, the China National Computer Network Emergency Response Technical Team (CNCERT) warned of prompt injection attacks, misinterpretation of user intents, malicious plugins, and exploitable security flaws that could spread misinformation or disrupt services.
These warnings highlighted that AI agents operate in dynamic, interactive environments where subtle security lapses can cause systemic risks. The Ministry of State Security’s (MSS) “Safe Lobster Farming Manual”, a reference to OpenClaw’s codename, addressed misinformation concerns and the need for comprehensive user and developer guidance. The manual’s viral popularity, topping one million views on platforms like RedNote/Xiaohongshu and widely shared by People’s Daily on WeChat, reflected broad public engagement and growing awareness about AI agent safety.
Hangzhou as the Vanguard of Autonomous AI Governance Trials
Amid this regulatory ferment, Hangzhou has become a strategic testing ground for autonomous AI governance models. Leveraging advanced digital infrastructure and strong government-industry collaboration, the city pilots frameworks for real-time monitoring, risk assessment, and automated compliance enforcement for AI agents within its jurisdiction. This pioneering approach combines local oversight with national policy to address challenges posed by AI agents’ operational autonomy.
Hangzhou’s governance ecosystem integrates AI safety tools developed by the Shanghai AI Lab, led by Zhou Bowen, chair of the AI Safety Working Group (WG9) convened by TC260. WG9 focuses on AI application security classification, security capability maturity assessments, and safety guardrails tailored for anthropomorphic interactive services. The city’s pilot programs also emphasize collaboration with cloud service providers, reflecting the AI Industry Alliance’s (AIIA) view of cloud platforms as critical governance levers due to their control over deployment environments and data flows.
The Complexity of AI Agent Regulation: Data Access and Liability Challenges
Regulatory challenges extend beyond technical security to complex issues of data governance and legal liability. As AI agents become more autonomous and integrated into daily life, regulators must balance agents’ need for broad, contextual user data access with privacy and security safeguards.
China’s policymakers debate how to define and enforce data usage boundaries to prevent misuse while enabling innovation. Liu Liehong, Director of the National Data Administration, emphasized that AI agents should evolve from “flashy all-capable executors” to “honest risk communicators and reliable solution providers,” implying design for transparency, ethical compliance, and trustworthiness.
Liability remains a thorny issue. Determining responsibility when an AI agent errs—whether developer, cloud provider, user, or the agent itself—is unresolved. The autonomous nature of agents blurs traditional accountability lines, prompting regulators to rethink legal frameworks and enforcement. Current debates focus on nuanced liability models reflecting AI-mediated interactions and distributed agent ecosystems.
Institutional Responses: Standardization and Cross-Sector Collaboration
In response, China’s AI regulatory ecosystem is rapidly maturing. TC260’s establishment of WG9 on March 20 formalizes AI safety governance, with an agenda including an AI security capability maturity model and frameworks for agent safety, anthropomorphic AI interactions, and data protection. The China Academy of Information and Communications Technology (CAICT) has expanded the “AI Safety Benchmark 2.0” to include frontier risks and scenario-specific safety metrics, moving toward more granular, context-aware evaluation standards.
China is drafting industry and national standards for the Model Context Protocol (MCP) security and agent security frameworks to create interoperable, enforceable guidelines for AI agent deployment. This standardization aims to harmonize technical security requirements across developers, cloud providers, and users, facilitating unified risk mitigation while encouraging innovation.
The AI Industry Alliance’s engagement of 13 leading Chinese companies in signing AI agent safety commitments highlights growing cross-sector collaboration. By positioning cloud providers as key governance actors, these initiatives recognize infrastructure operators’ critical role in enforcing security policies and managing data flows, broadening the regulatory perimeter beyond developers to include service ecosystems.
Implications for China’s AI Industry and Global Positioning
China’s evolving AI agent regulation marks a strategic pivot in managing emerging technologies with profound social and economic impacts. By addressing AI agent risks through integrated governance—combining technical standards, legal frameworks, and local experimentation—China aims to balance innovation leadership with societal trust and security.
Hangzhou’s autonomous governance experiments model scalable AI oversight adaptable to rapid technological change and complex risks. This decentralized yet coordinated approach may give China a competitive edge in managing AI deployment at scale amid intensifying international AI regulation debates.
China’s focus on agent safety and liability preempts incidents that could erode public confidence or provoke external regulatory backlash. Embedding safety guardrails and accountability early positions China to maintain control over AI’s societal impact while fostering a robust domestic AI ecosystem.
Geopolitically, these regulatory developments signal China’s commitment to shaping AI governance’s future. As AI agents grow more capable and ubiquitous, China’s emerging frameworks will influence international norms, standards, and cross-border cooperation on AI safety.
Toward a New Paradigm of AI Agent Safety in China
China’s AI agent regulation debate has entered a new era marked by heightened government engagement, institutional innovation, and experimental governance models. The proliferation of AI agents like OpenClaw exposed critical vulnerabilities, prompting a sophisticated policy response integrating security standards, liability frameworks, and data governance into a cohesive strategy.
Hangzhou’s role as a testing ground for autonomous AI governance illustrates China’s pragmatic approach—leveraging local innovation to refine national policy. The formation of specialized working groups and expanded safety benchmarks demonstrate a maturing regulatory ecosystem addressing AI agents’ unique challenges.
EastFrontier has recently reported on other areas of AI safety in China, including the digital human industry and its ban on virtual companions for minors. China’s experience offers valuable insights into the interplay of technology, policy, and society in AI autonomy’s age. These efforts will shape China’s AI industry trajectory and the global discourse on safely governing increasingly intelligent and independent AI systems.