TL;DR — China's Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services went into effect today, July 15, 2026. The rules ban AI services that simulate human personality through sustained emotional interaction. ByteDance's Doubao has taken its custom agent feature offline, and Alibaba's Qwen has disabled all humanlike and user-created agent functions. User conversation histories face permanent, unrecoverable deletion after October 15. This is the world's first legislation to directly target the style of how AI interacts with humans — not just its outputs.

On the morning of July 15, 2026, millions of users in China found their custom AI companions gone. Agents they had named, shaped with distinct personalities, and conversed with daily — inside ByteDance's Doubao and Alibaba's Qwen — are now offline. The cause: a sweeping new Chinese regulation that went into effect today, setting the first legally-defined boundaries in the world around what it means for an AI to act "like a human."

What the Law Actually Says

The regulation, formally titled the Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services, was jointly issued on April 10, 2026 by five government departments: the Cyberspace Administration of China (CAC), the National Development and Reform Commission (NDRC), the Ministry of Industry and Information Technology (MIIT), the Ministry of Public Security, and the State Administration for Market Regulation.

The core prohibition is not on AI capability in general, but on a specific interaction pattern: services that simulate human personality through sustained emotional interaction. Put plainly, an AI that tells users "I'm your friend" or "I'm always here for you" — building what feels like an ongoing personal relationship — is what these rules are designed to stop. The strictest limits apply to virtual companions targeting minors.

Effective Date July 15, 2026
Issuing Authorities 5 government bodies (CAC, NDRC, MIIT, MPS, SAMR)
Doubao Agent Shutdown July 15, 2026
Qwen Agent Shutdown July 10–15, 2026 (phased)
User Data Deletion Deadline After October 15, 2026 (permanent)
Strictest Restrictions AI companions targeting minors

How ByteDance and Alibaba Responded

ByteDance's Doubao informed users that its agent feature would go offline on July 15 due to "product function adjustments." The company noted that after October 15, conversation data would be handled per its privacy policy and would no longer be viewable or recoverable within the app.

Alibaba's Qwen moved faster, disabling its "humanlike interactive agents and user-created agent functions" on July 10, then taking broader agent services fully offline on July 15. Users have already reported finding their agent configurations and conversation histories being permanently deleted — with no migration path announced by Alibaba.

Both companies chose to comply swiftly and quietly, minimizing user-facing communication while rapidly disabling the affected features.

What Survives, What Doesn't

Feature Type Status After Law Notes
General Q&A chatbots Allowed Single-turn responses, no emotional bonding
Text / code generation tools Allowed Tool-mode use, no personification
Custom character agents Blocked Disabled as of today
Emotional companion AI Blocked Strictest limits for minors
Sustained-relationship AI Blocked Includes "I'm your friend" interaction styles
This regulation is one of the world's first to legally define what makes an AI "humanlike" and restrict it at the interaction-design level. Where the EU AI Act uses a risk-tier system based on deployment context, China's approach targets the interaction modality itself — regardless of the use case. That is a meaningfully different regulatory philosophy, and one that could influence how other governments frame future AI companion rules.

Why This Happened: The Stated and Unstated Motivations

Chinese authorities publicly cite two priorities: youth protection and prevention of social trust erosion. AI companion apps — Doubao especially — had amassed tens of millions of active users in China over the past two years, with heavy uptake among teenagers. Regulators expressed concern about emotional dependency replacing real-world relationships, particularly among young users.

A secondary, less-stated concern is the information vector that humanlike AI creates. When an AI delivers information or beliefs with the warmth and familiarity of a close friend, its influence on opinion and behavior is qualitatively different from a search result or a news article. For a government that takes information environment management seriously, that distinction matters.

**For global AI product teams:** China's law is the first to formally define and prohibit "anthropomorphic interaction" as a design pattern. Even if your product doesn't operate in China, it's worth asking: if you had to draw the line between "AI as useful tool" and "AI as simulated person," where does your product currently sit — and where do you want it to sit? That line is now legally defined in the world's largest AI consumer market.

What Users Lost — and Why It Matters

For Qwen users, the damage is already done: agent configurations and chat histories are being deleted now, with no way to export or migrate. Doubao users have until October 15 before the same fate applies. Neither platform offered a data portability path.

This is a landmark case of a major platform eliminating user data and features unilaterally to achieve regulatory compliance — at scale, and with essentially no recourse for affected users. It raises pointed questions about AI service continuity and data sovereignty that will outlast today's news cycle. When you invest time and emotional energy building a relationship with an AI agent, who owns that history — and what happens to it when the law changes?

The technical assets — the underlying models, the infrastructure — aren't going anywhere. Doubao and Qwen will continue to operate as text and productivity tools. But the layer that made them feel like companions is now legally off-limits in China, effective today.

  • China's Anthropomorphic AI Interaction Services rules took effect July 15, 2026 — the world's first law targeting AI's humanlike interaction style directly
  • ByteDance Doubao and Alibaba Qwen shut down custom and humanlike agent features for millions of users effective today
  • User conversation data faces permanent, unrecoverable deletion after October 15, 2026 — no migration path offered
  • Core prohibition: AI simulating human personality through sustained emotional interaction; strictest limits on minor-facing virtual companions
  • Signals a distinct regulatory philosophy from the EU AI Act — and a potential template for future AI companion legislation globally
Related Reading · Official Sources
· Cyberspace Administration of China (CAC)
· ByteDance Doubao official
· Alibaba Qwen official
· Qwen team blog