The regulation — co-issued in April by the Cyberspace Administration of China and four partner agencies — draws a deliberate line between the agent that does your work and the agent that keeps you company. Customer service bots, workplace assistants, and education tools are excluded. What it targets is services that simulate human personality traits and communication styles to sustain emotional interaction: the named companions, role-play characters, and therapy-style personas that millions of users built on China's biggest consumer AI apps.
The compliance demands — anti-addiction systems, mandatory usage notifications, instant-exit mechanisms, real-time detection of unhealthy dependence, and a full ban on virtual companions for minors — sit awkwardly with agents designed to remember users and maintain ongoing relationships. Rather than retrofit, ByteDance and Alibaba pulled the features. Doubao's agents went offline July 15 with read-only access until October 15, after which the data becomes unrecoverable. Qwen disabled humanlike agents on July 10 and broader agent services on July 15, with configurations and chat histories set for permanent deletion and no announced migration path. Tencent's Yuanbao removed a comparable feature in June, and Shanghai's cyberspace authority had already purged more than 14,000 non-compliant agents in a pre-deadline sweep.
The user cost is real: people mourned the shutdowns on Weibo, describing agents that had served as long-standing emotional support. ByteDance is steering users toward Maoxiang, a separate app built for compliance; Alibaba has offered nothing equivalent. The pattern is a signal for the global market — California's SB 243 took effect in January, and the EU is moving in the same direction.
