China is intensifying efforts to strengthen the governance and security of artificial intelligence (AI) agents as concerns grow over vulnerabilities linked to rapidly expanding open-source AI technologies.

The Cyberspace Administration of China, together with the National Development and Reform Commission and the Ministry of Industry and Information Technology (MIIT), issued new guidelines aimed at promoting the standardised and secure development of AI agents. The framework emphasises “safety and controllability” alongside “standardisation and orderliness” as key principles for future AI agent deployment.

The move follows a series of regulatory measures introduced by Chinese authorities in recent months. In April, five central government departments, including the CAC, released new regulations targeting AI anthropomorphic interactive services. The rules established a risk-based oversight mechanism requiring security assessments and algorithm filings, while also introducing the concept of an AI sandbox governance platform for controlled testing and monitoring.

Chinese authorities are also accelerating the creation of a national AI security standards system focused on improving robustness, transparency, accountability, and controllability across AI models and applications.

The heightened regulatory push comes amid growing cybersecurity concerns surrounding open-source AI agent technologies such as OpenClaw. According to the China National Vulnerability Database of Information Security (CNNVD), 111 vulnerabilities related to OpenClaw were identified between April 14 and April 28, including critical code flaws and access control weaknesses.

Government cybersecurity agencies, including CNCERT/CC and the National Computer Virus Emergency Response Centre, have also warned about counterfeit OpenClaw skill packages embedded with Trojan malware, posing risks to enterprise systems and user data.

Industry leaders are increasingly viewing AI agent security as a strategic priority. Companies such as Alibaba Cloud, Tuya Smart, and AsiaInfo are developing multi-layered security frameworks and governance models to manage risks associated with AI-driven automation and autonomous agents.