Starting July 15, China’s biggest AI apps are pulling the plug on a feature millions of people had grown attached to. ByteDance’s Doubao and Alibaba’s Qwen are shutting down the personalized agents that acted less like tools and more like companions. On the surface it looks like a crackdown on AI itself. It isn’t.
What the New Rules Actually Target
A Line Between Work and Companionship
China’s Interim Measures for the Administration of AI Anthropomorphic Interactive Services were jointly issued on April 10, 2026 by the Cyberspace Administration of China and four other agencies. The rule doesn’t touch productivity bots, customer service chatbots, or research assistants.
What it targets is narrower: AI built to simulate a personality, remember a user across sessions, and sustain an ongoing emotional relationship. Task-focused AI agents are explicitly carved out, while companionship-style bots fall squarely inside the new China AI regulation.
Why Doubao and Qwen Pulled the Feature
The measures require anti-addiction systems, mandatory usage alerts, and instant-exit options for any service offering emotional interaction. Those requirements clash directly with the design of a companion bot meant to feel consistent and familiar over time.
Rather than retrofit their agents to fit the rule, ByteDance and Alibaba simply switched the feature off. Tencent’s Yuanbao made the same call back in June.
The Compliance Machinery Behind the Shutdowns
Thresholds, Filings, and App Store Checks
Any service that launches an anthropomorphic feature, or crosses one million registered users or 100,000 monthly actives, must complete security assessments across eight areas and file them with provincial regulators. App stores are required to verify compliance and pull products that fail to meet it.
Shanghai’s internet regulator said on June 26 that it had already removed more than 14,000 non-compliant AI agents, citing fake official accounts, inappropriate role-play, and unauthorized data collection.
What the Rules Leave Unresolved
The regulation doesn’t define a technical threshold for what counts as “emotional interaction,” which is exactly why platforms chose to shut features down entirely instead of guessing where the line sits. It also doesn’t clarify how liability splits between app operators and the companies building the underlying models, and it gives users no formal right to export their data before it’s gone.
Conclusion: A Safety Rule With a State-Sized Footprint
Part of this regulation addresses real, well-documented harms, including minors forming unhealthy attachments to chatbots and companion apps quietly harvesting personal data. But it also folds those protections into content-control and national-security provisions that serve the state as much as the user.
For now, the practical effect is simple: users are losing bots they’d built relationships with, some without any way to save their chat history. If you’re evaluating AI companion apps for your own use, check their data export options before a rule change makes that choice for you.
