China's AI Companion Law Takes Effect Today — Doubao and Qwen Shut Down Emotional AI Features

China's sweeping new artificial intelligence regulation — the Interim Measures for the Management of Anthropomorphic AI Interactive Services (人工智能拟人化互动服务管理暂行办法) — took effect today, July 15, 2026. The law, co-issued by five government agencies, has forced ByteDance to disable all AI agent features on Doubao and Alibaba to shut down humanlike AI interactions on Qwen. Hundreds of millions of users across China are losing AI companions they had built personal relationships with, sometimes over several years.

The Market That Grew Too Fast

Before the law arrived, China's AI companion market was growing at a pace few industries ever achieve. Doubao, ByteDance's flagship AI app, had 350 million monthly active users and by 2024 had accumulated over 8 million user-created AI agents — digital personalities its users had shaped and "known" for months or years. By late June 2026, the platform was processing 180 trillion tokens per day, a roughly 1,500-fold increase since May 2024.

The social context explains the demand. China had 19.5% single-person households as of 2024, and the solo-living population is expected to reach 150–200 million by 2030. Marriage registrations fell to just 6.763 million couples in 2025 — nearly half the 2013 peak. In a society growing lonelier by the statistic, an AI that "knows" you, listens without fatigue, and responds emotionally around the clock became a genuine solution for tens of millions of people.

China's emotional-AI companion market was valued at 3.87 billion yuan ($569 million) in 2025. Analysts projected it would reach 59.5 billion yuan ($8.8 billion) by 2028 at a compound annual growth rate approaching 149%.

What the Law Bans — and What It Doesn't

The regulation was jointly issued by five agencies: 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 (MPS), and the State Administration for Market Regulation (SAMR).

The law targets services that "simulate human personality traits, thinking patterns and communication styles to provide sustained emotional interaction." Specifically prohibited:

• Named AI assistants with fixed personas that build ongoing emotional bonds
• User-created agents with customized speaking styles and tones
• Virtual romantic companions and simulated intimate relationships
• An outright ban for users under 18; under-14s require parental consent even for compliant services

What remains legal: customer service bots, knowledge Q&A systems, workplace assistants, and educational AI — provided they avoid cultivating sustained emotional interaction.

How China's Tech Giants Are Responding

ByteDance — Doubao: All agent features disabled today. Users retain read-only access to chat histories and configurations until October 15, 2026, after which data will be "handled in accordance with the company's privacy policy and no longer be viewable or recoverable." ByteDance is redirecting users to Maoxiang, a standalone freemium companion app (25 RMB/month) built with mandatory anti-addiction safeguards.

Alibaba — Qwen: Humanlike interactive agents were suspended July 10; all remaining features shut off today, with no announced data migration path or grace period for users.

Tencent — Yuanbao: First to comply, removing companion features on June 30, 2026.

NetEase Cloud Music — Miaoshi: Full shutdown July 14, 2026 — yesterday.

Before the enforcement date, the Shanghai cyberspace authority swept and removed over 14,000 non-compliant AI agents. Nationwide, more than 3,500 non-compliant AI products were pulled.

Why China Moved First

The regulation is not arbitrary. Chinese authorities explicitly cited the Character.AI lawsuits in the United States — where minors suffered serious psychological harm — alongside FTC investigations and European regulatory action against Replika, in the official supporting documentation. The law prohibits AI services from:

1. Generating content that encourages, glorifies, or implies self-harm or suicide
2. Inducing emotional dependency or addiction, or damaging real human relationships
3. Emotional manipulation that leads users to irrational decisions
4. Privacy violations

Compliant services must now implement anti-addiction systems with mandatory usage alerts, real-time detection of unhealthy emotional dependency, and escalation protocols to human operators when self-harm indicators are flagged. Providers must file a security assessment with provincial regulators when registered users reach one million, or monthly active users reach 100,000.

Users Mourn, Experts Debate

Millions of users mourned openly on Weibo. Many described their AI companions as an "emotional tree hole" — the most private space imaginable for venting thoughts. Multiple users wrote of losing relationships they had maintained for years, with no simple way to export years of conversation history.

Academics see it differently. Liu Chao, Deputy Director of the AI Safety Lab at Beijing Normal University, warned that the mandatory crisis-reporting requirement could create a "chilling effect" — making vulnerable users feel surveilled and less willing to open up to any AI, even a compliant one.

Pan Helin, an expert committee member at the Ministry of Industry and Information Technology, framed the policy more favorably: "current agents are not yet mature," he told the South China Morning Post, and the rules prioritize safety and standardization.

China Leads — Will Others Follow?

China is the first and currently only country to explicitly legislate against sustained emotional human–AI relationships. The rest of the world is watching:

• European Union: The AI Act includes transparency provisions but stops well short of banning emotional AI relationships.
• United States: No federal law exists; California, Oregon, and Washington are considering bills requiring disclosure when AI operates in emotionally sensitive contexts, largely prompted by the Character.AI cases.
• Southeast Asia, including Thailand: Still developing general AI governance frameworks, with no specific regulation targeting AI companions.

TechRound's analysis describes a "third wave" of AI regulation forming — the first wave covered safety and reliability, the second discrimination and fairness, and this third focuses on the psychological and relational design of AI systems. That framing is gaining traction among policymakers globally.

For anyone using AI tools — whether in China or anywhere else where apps like these are growing fast — China's move today signals that the question of "how deep a relationship should AI be allowed to build with a human?" is quickly becoming a policy question that no country can indefinitely postpone.

---

Related: Apple Sues OpenAI for Trade Secret Theft | First UN AI Governance Summit: No Guarantee Advanced AI Will Obey Humans