On July 15, 2026, China became the first major AI market to switch off an entire product category. The Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services — Beijing’s new rules for human-like AI — took effect, and the country’s two biggest consumer AI apps complied in the most visible way possible: ByteDance’s Doubao and Alibaba’s Qwen shut down their customizable agent features, wiping out millions of user-created AI companions overnight.
⚡ Key Facts
- The rules took effect July 15, 2026, after being co-issued on April 10 by the Cyberspace Administration of China and four other agencies
- They target bots that offer sustained emotional interaction by simulating human personality, thinking patterns, and communication styles
- Workplace and productivity agents are explicitly spared — this is a companion-bot law, not an agent ban
- Doubao’s agent feature went offline July 15; Qwen disabled human-like and user-created agents on July 10
- Doubao users have until October 15 to export their data; Qwen offered no migration at all
What the Regulation Actually Says
The Interim Measures were co-issued by five agencies — the Cyberspace Administration of China (CAC), 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. That lineup matters: when five regulators sign one document in China, enforcement is not optional.
The scope is narrower than early headlines suggested. The rules cover services that simulate human personality traits, thinking patterns, and communication styles to provide sustained emotional interaction — in plain terms, AI companions, virtual girlfriends and boyfriends, digital personas, and role-play agents. What they deliberately leave alone is the productivity side: coding assistants, workplace copilots, customer-service bots, and task agents remain untouched.
How Doubao and Qwen Complied
Both companies moved before the deadline rather than after it, which tells you how seriously Chinese platforms take CAC effective dates.
📅 Shutdown Timeline
| Date | What happened |
|---|---|
| April 10, 2026 | Five agencies co-issue the Interim Measures |
| July 10, 2026 | Qwen disables human-like interactive agents and user-created agents |
| July 15, 2026 | Rules take effect; Doubao’s agent feature goes offline; Qwen’s broader agent services follow |
| October 15, 2026 | Deadline for Doubao users to export their companion data — Qwen offers no export at all |
Doubao framed the shutdown to users as “product function adjustments” — corporate language for a regulatory kill switch. The user impact is real: people who spent months building customized companions lost them with a few days’ notice, and in Qwen’s case, without any way to take their conversation history with them.
Why Beijing Is Doing This
The regulation reflects a concern Chinese authorities have voiced repeatedly: emotional dependency on synthetic personas, particularly among minors and lonely users. By carving out productivity agents while banning sustained emotional simulation, Beijing is drawing a line between AI as a tool — which it is aggressively subsidizing — and AI as a relationship, which it now treats as a social risk.
There is also an industrial-policy angle. China’s AI champions — the same companies behind DeepSeek, Qwen, and GLM — are being steered toward enterprise value and away from consumer engagement farming. Companion bots monetize attention; Beijing wants its compute spent on capability.
💡 Convly’s Take
This is the sharpest regulatory divergence in AI so far. Western labs ship increasingly personable assistants with no comparable restrictions, while Chinese platforms just deleted the category. Two things follow. First, companion-AI innovation will migrate to Western and gray-market apps — and so will its problems. Second, Chinese consumer AI apps lose their stickiest engagement surface, which pushes ByteDance and Alibaba even harder toward the enterprise-agent race where the rules leave them room. Watch whether other regulators treat Beijing’s carve-out — tools yes, relationships no — as a template.
What It Means for the Global AI Race
For Western users and builders, nothing changes today — and that is precisely the point. OpenAI, Anthropic, Google, and Meta face no equivalent of the Interim Measures. If anything, the shutdown creates a natural experiment: one market with emotional AI, one without. Researchers will be citing this split for years.
For Chinese users, the productivity carve-out means the models themselves are not going anywhere. Qwen remains one of the strongest open-model families in the world, and Doubao remains ByteDance’s consumer AI flagship — just with the personality dialed down.
📚 Go Deeper on Chinese AI
- ByteDance Doubao, explained — China’s biggest consumer AI app
- Alibaba Qwen, explained — the open-model powerhouse
- DeepSeek, explained — the efficiency king of our price-performance index
- Moonshot Kimi, explained
FAQ
Does China’s ban affect AI models like Qwen or DeepSeek themselves?
No. The rules target human-like interaction services — companion and role-play agents — not the underlying models. Qwen’s open-weight models remain available, and productivity uses are explicitly spared.
Can Doubao users recover their AI companions?
No. The feature is offline as of July 15, 2026. Users can export their data until October 15, 2026, but the companions themselves are gone. Qwen users received no export option.
Do similar rules exist outside China?
Not yet. No Western jurisdiction currently bans emotional-companion AI, though several are studying child-safety rules for chatbots. China’s Interim Measures are the first enforced, category-wide restriction.
