China AI Companion Rules Hit Doubao and Qwen
2 min readChina’s new rules on humanlike AI took effect on July 15, and the country’s biggest chatbots are already pulling features to comply. ByteDance’s Doubao and Alibaba’s Qwen have begun disabling the custom, humanlike AI agents that let users build companions with distinct personalities.
What the Rules Require
The regulation is formally called the Interim Measures for the Administration of AI Anthropomorphic Interactive Services. The Cyberspace Administration of China co-issued it in April alongside four partner bodies, including the Ministry of Industry and Information Technology and the Ministry of Public Security. It targets services built around sustained, humanlike emotional interaction, while deliberately sparing workplace and productivity agents.
Under the measures, companion services must run anti-addiction systems, send mandatory usage notifications, offer an instant-exit option, and detect signs of unhealthy dependence in real time. Beijing’s stated concern is that emotionally lifelike bots can foster attachment, especially among younger or vulnerable users.
How the Platforms Responded
Doubao, China’s most popular AI chatbot, told users its agent feature would go offline on July 15 due to “product function adjustments,” according to the South China Morning Post. Qwen issued a similar notice, disabling user-created agents and broader agent functions on the same timeline. Tencent’s Yuanbao and other major platforms have reportedly made comparable changes.
The data consequences are significant. Doubao said related agent data would become unrecoverable inside the app after October 15. Qwen users were given no equivalent grace period, meaning custom character configurations and conversation histories will be permanently deleted with no announced migration path.
Why It Matters
AI companions have surged in popularity across China, and shutting them down almost overnight shows how quickly Beijing can reshape a fast-moving consumer market. The rules also rank among the world’s first concrete attempts to regulate emotional AI, an area Western regulators are only beginning to study.
For ByteDance and Alibaba, the message is that anthropomorphic features now carry real compliance risk. Expect other Chinese firms to steer their AI toward tools and assistants, and away from anything that feels too much like a friend.