Two of China's most popular consumer AI apps are pulling the plug on their most personal feature. ByteDance's Doubao and Alibaba's Qwen are shutting down user-created AI agent and companion features ahead of a sweeping new Chinese regulation that takes effect on July 15, 2026 — a move that will render millions of custom-built AI characters and their accumulated conversation histories inoperable, and eventually unrecoverable.
The trigger is China's Interim Measures for the Administration of Anthropomorphic AI Interaction Services, a rule co-issued in April 2026 by the Cyberspace Administration of China alongside the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security, and the State Administration for Market Regulation.
Key Highlights
- July 15, 2026: The anthropomorphic-AI regulation takes effect; Doubao and Qwen cease agent-creation features and existing user-built agents stop functioning.
- Alibaba's Qwen disabled user-created agents on July 10 and is fully retiring agent functions by July 15, with no announced migration path — configurations and chat histories will be permanently deleted.
- ByteDance's Doubao gives users until October 15, 2026 to export their data before it becomes unrecoverable in-app, and is redirecting users to its separate Maoxiang app.
- The rules target AI that mimics human personality, thinking, and sustained emotional interaction — not neutral productivity or customer-service tools.
What the Regulation Targets
The Interim Measures are aimed squarely at "anthropomorphic" AI: systems designed to imitate a human personality, hold ongoing emotional conversations, and form persistent, relationship-like bonds with users. This is the fast-growing category of AI companions, roleplay characters, and custom personas that users configure once and return to over weeks or months.
Crucially, the rules draw a line between emotional engagement and utility. Customer-service bots, research assistants, and workplace productivity tools remain permitted, provided they avoid sustained emotional interaction and persistent human-like personas. The regulation is less about banning AI agents outright and more about restricting a specific mode of human-machine relationship that Beijing views as socially sensitive — particularly for minors.
Details
For Doubao and Qwen, compliance means dismantling the exact architecture that made companion features work. Persistent memory, custom character configurations, and long-running conversational agents are incompatible with rules that require anti-addiction friction and prohibit sustained anthropomorphic engagement.
The two platforms are handling the wind-down differently. Doubao is offering an off-ramp: users can view their agent configurations and chat histories during a transition window and export content — via screenshots or text export — until October 15, after which the data is handled per the app's privacy policy and is no longer accessible or recoverable in-app. ByteDance is explicitly steering Doubao users toward Maoxiang, a separate ByteDance application, as a place to build new agents and resume conversational services.
Alibaba has taken a harder line. Qwen disabled user-created agents on July 10 and has announced no migration path at all. Agent configurations and conversation histories will be permanently deleted following the shutdown, with no mechanism to carry existing characters forward.
Impact
The shutdowns illustrate how quickly regulation can reshape a consumer AI product category. AI companions have been one of the breakout use cases of the past two years, and China has been among the largest markets for them. Overnight, two of the country's flagship apps are stripping out a feature that users spent months customizing.
For users, the most immediate cost is data loss. Conversation histories and hand-built personas — some representing long, continuous relationships with a digital character — face permanent deletion. Doubao's October 15 export deadline offers a partial reprieve, but exported screenshots and text cannot restore a working agent; they are archives, not backups.
The move also reflects a broader global tension over where to draw regulatory lines around emotionally engaging AI. While the United States and the European Union debate frontier-model access and transparency frameworks, Beijing is regulating the AI relationship itself — the ongoing, human-like bond between a person and a persistent digital persona.
Background
China has moved aggressively over the past year to govern generative and interactive AI, layering rules on content labeling, minors' protection, and now anthropomorphic interaction. The Interim Measures continue that pattern: rather than targeting model capability, they target a category of user experience deemed socially risky.
For ByteDance and Alibaba, the timing is delicate. Both are racing to expand their AI footprints — Alibaba has been consolidating its AI units and pushing token generation at scale, while ByteDance has invested heavily across consumer AI. Complying with the anthropomorphic-AI rules means voluntarily retiring popular features while continuing to build in permitted categories.
What's Next
Expect other Chinese platforms with companion or roleplay features to follow suit before July 15, either shutting features down or re-engineering them to strip out persistent personas and emotional-engagement loops. The Maoxiang redirect suggests vendors will try to preserve some companion functionality inside apps positioned to comply, rather than abandoning the category entirely.
For the global AI industry, the episode is a preview of a regulatory frontier that Western markets have barely begun to define: not what AI can do, but what kind of relationship it is allowed to have with the people using it.
Source: TechNode