AIBRIEF
Two voluntary Chinese industry rules, drafted by a consortium led by the China Cyberspace Security Association with Ant Group, govern an AI agent's identity and runtime behavior.
China's first two AI agent security group standards launched on September 9, 2026 at the Inclusion·Bund Conference in Shanghai. They govern "AI agents", software entities that plan and act on their own behalf, calling tools and APIs, not just answering prompts. One standard handles who the agent is and what it can do, the other monitors what it actually does at runtime, from intent to tool call to result.
The first standard, 《智能体身份鉴别与授权技术框架》, builds a multi-layer trusted identity chain between agent, user, and runtime environment, and adds task-scoped access control: permissions granted on task start, revoked on completion. The second, 《智能体运行时安全技术要求》, covers runtime monitoring, intent-drift detection, and defense-in-depth across the full execution path: user intent → agent planning → tool call → system execution → result delivery.
Both are group standards (团体标准): voluntary, industry-led technical rules, not government regulation. The China Cyberspace Security Association proposed them; Ant Group, research institutes, telecom operators, and internet vendors co-drafted. At the forum, Ant Group VP of Big Security Wang Yu called agents a new "subject" in the digital world whose identity, authority, intent, and behavior must all be auditable. Identity is the foundation, runtime the enforcement. That is the consortium's pairing logic.
The launch lands inside a wider policy track: the CAC's May 2026 implementation opinion on agent application, the TC260 deployment-and-use security practice guide, the SAMR national-standard project on basic agent-application security requirements, and a July 2026 Xinhua report on China's first mandatory national agent-security standard. The first adoption test is whether the SAMR mandatory track picks up the same identity-then-runtime pairing. If it does, the group standards become the template.