China launches first policy framework for AI agents alongside companion rules

A July 2026 issue of the China AI Law Brief, written by Sarah Zhao, a partner at Rimon Law, analyses three Chinese AI developments taking effect that month. The author sees a shift from broad principles to operational rules, through one voluntary guideline and two binding instruments. Their subjects are AI ethics, agents, and human-like interactive AI services.

The brief explains the background. Open-source agent technology, notably OpenClaw, spread quickly in late 2025 and early 2026, and reported incidents involved credential theft through malicious plug-ins, data leakage and prompt injection. Meanwhile AI companion products grew quickly, raising concerns about emotional dependence among minors and elderly users, and existing rules such as the Interim Measures for Generative AI Services gave limited guidance on these areas.

On agents, the Implementation Opinions on Intelligent Agents, covering their standardized use and innovative growth, issued jointly by the CAC, NDRC and MIIT and effective 15 July 2026, create what the brief calls China's first dedicated policy framework for agents as a category separate from generative AI. The text defines an agent as a system able to perceive, remember, decide, interact and act on its own. Agents deployed in sectors such as health, transport, media and public safety must meet compulsory filing, compliance testing and recall rules. The Opinions identify nineteen application scenarios and target 70% adoption of agents in smart terminals by 2027.

Alongside this, the interim measures on anthropomorphic AI interaction services, also effective 15 July, require algorithm filing, security assessments, disclosure that content is AI-generated, addiction prevention and limits on virtual intimate relationships for minors. The Ethics-Safety Guidelines for AI Applications 1.0, effective 1 July, set nine ethics principles. The author advises companies to review their products against all three.