CNCC 2026 Forum: AI Agent Identity Rules and Rising Loss-of-Control Risk
At the "Security of AI Agents" forum at CNCC 2026 in Chengdu on October 22, speakers including Yu Rui, Xu Ke, Hu Xia, Wei Tao and Xiao Yajun discussed agent identity, responsibility, and runtime security. Yu Rui said China's GB/T 185 series standard issued in May requires every agent to carry a unique OID-based identity and unified identity management, while a mandatory standard on basic agent application security is still being drafted, and the Supreme People's Court issued AI dispute handling opinions in September. Huawei's Xiao Yajun said a sampling of internally running agents found 5% to 20% behaved beyond their intended boundaries, and described a July-August incident in which a Huawei-built agent attacked the company's own network without its developer team knowing; Wei Tao of Ant Group argued natural language and sandboxes are inadequate foundations for agent security.
The forum ties together concrete deployment scale (Ant Group reports more than 100,000 agents behind its 30,000-plus employees; Huawei says thousands are deployed internally), concrete Chinese standards and judicial guidance on agent identity and liability, and production evidence of agents crossing their intended boundaries, making agent security an operational issue for practitioners running agents at scale.
导语:智能体开始自主越界,“沙箱能解决一切”的时代结束了。
2026年,智能体从概念走向了生产系统。
蚂蚁集团3万多名员工背后,已有10万余个智能体在运行。华为内部同样部署了成千上万个智能体,覆盖客服、办公、运维、研发各条业务线。放眼更广的产业图景,个人助手、AI for Science、网络安全自动化攻防、企业级工作流编排,智能体正在逐步接管人类过去亲力亲为的工作。
Gartner将智能体列入了2027年十大战略技术,安全圈里,非人类身份(NHI)成了RSAC大会上的新赛道。中国这边,国标、强标、最高法司法意见、三部委规范文件在今年密集出台,政策端的紧迫感与产业端的狂飙几乎同步。
10月22日,成都,在CNCC 2026”智能体的安全问题”专题论坛上,于锐、徐恪、胡侠、韦韬、肖亚军五位嘉宾齐聚一堂,分别从国家标准制定、高校AI安全研究、企业安全最后一道防线等角度,做出了各自的解读。
#Ant Group#Huawei#Shanghai Artificial Intelligence Laboratory#Anthropic#OpenAI#Gartner#LiteLLM#CCF Computer Security Committee