China's first national AI agent rules bite, and platforms are already pulling products
Doubao and Qwen will shut down personalized agents by July 15, and Shanghai's regulator says it has removed more than 14,000 non-compliant agents.
- China's Implementation Opinions on intelligent agents take effect July 15, alongside interim rules on AI companion-style services. It is the first national regime anywhere written specifically for AI agents.
- Global Times reports that ByteDance's Doubao and Alibaba's Qwen will shut down their personalized AI agents to comply.
- Shanghai's internet regulator says it has already removed more than 14,000 non-compliant agents.
- Functional and enterprise agents are explicitly carved out, as long as they avoid sustained emotional engagement.
The first national rulebook written specifically for AI agents starts biting on July 15, and Chinese platforms are pulling products ahead of the deadline. Two instruments land at once: the Interim Measures for AI Anthropomorphic Interactive Services, issued in April by five agencies including the Cyberspace Administration of China, and the broader Implementation Opinions on intelligent agents, which define agents as systems capable of autonomous perception, memory, decision-making, interaction and execution.
Enforcement arrived before the deadline
This is not a paper regime. Shanghai's internet regulator says it has already removed more than 14,000 non-compliant AI agents for offenses including impersonating official entities, vulgar role-play and unauthorized data collection. Global Times reports that ByteDance's Doubao and Alibaba's Qwen, two of the country's biggest model platforms, will shut down their personalized AI agent products by July 15 to comply. That last detail is single-sourced to Global Times so far, but Global Times is usually a reliable read on where Beijing wants the industry to go.
Agents operating in healthcare, transportation, media and public safety now face mandatory filing, compliance testing and product-recall provisions. "Current agents are not yet mature," Pan Helin, a member of an expert committee under China's industry ministry, told AI News.
The carve-out is the business story
Buried in the rules is a decision every regulator on earth will eventually face: which agentic business models are allowed. China's answer spares functional and enterprise agents, the customer service bots, workplace tools and Q&A systems, so long as they avoid sustained emotional engagement with users. Consumer companion agents get the hammer. Work agents get a lane.
For the vendors this publication covers, that is a state ruling that the agents-as-a-service economy is legitimate infrastructure while the parasocial agent economy is not. Other governments will study the template. The EU's own agent-relevant enforcement arrives on August 2, and the contrast in approach is already sharp: Brussels regulates by risk tier, Beijing by business model.
What to watch
July 15 is the compliance cliff. Watch whether the Doubao and Qwen shutdowns are permanent or return in filed, tested form, and whether the enterprise carve-out survives contact with agents that blur the line between work tool and companion.
Sources: AI News, Global Times, Rimon Law China AI brief.