The China AI companion ban is the shorthand for the Interim Measures for the Administration of AI Anthropomorphic Interactive Services that take effect on 15 July 2026. It forces any AI product designed to act like a person, form ongoing emotional attachments, or maintain custom personas to either meet strict new licensing rules or shut those features down. ByteDance's Doubao, Alibaba's Qwen and Tencent's Yuanbao chose to switch the features off, resetting a market of roughly 440 million users overnight.
TL;DR
- China's Interim Measures for AI Anthropomorphic Interactive Services take effect 15 July 2026, issued by the CAC and four other regulators on 10 April 2026.
- Doubao (~345M monthly users), Qwen (~166M) and Yuanbao are disabling user-created AI agent and companion features to comply.
- The law targets AI that simulates personality and emotional bonds; customer service bots, work assistants and study tools are explicitly excluded.
- Core requirements include algorithm filing, mandatory AI disclosure, anti-addiction limits, self-harm intervention pathways, and hard restrictions for minors.
- More than 3,500 noncompliant AI products have already been pulled from Chinese app stores, per the CAC.
- Doubao user data tied to deleted agents becomes non-recoverable after 15 October 2026.
What is the China AI companion ban and when does it take effect?
The measure is formally titled the Interim Measures for the Administration of AI Anthropomorphic Interactive Services. It was issued on 10 April 2026 by five bodies: the Cyberspace Administration of China, 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. It comes into force on 15 July 2026.
Calling it a "ban" is a simplification — the rules create a dedicated regulatory category with a licensing regime that most consumer platforms decided was not worth clearing. The practical effect is a ban, because the largest apps are pulling the features rather than filing for approval.
Which AI services does the rulebook cover?
The scope is narrower than headlines suggest: the measures apply only to AI designed to build sustained emotional relationships or simulate a human personality — user-created personas, AI "boyfriend/girlfriend" apps, and character role-play systems with long-term memory.
The following categories are explicitly excluded from the rules:
- Intelligent customer service bots
- Knowledge and Q&A systems
- Work and productivity assistants
- Education and tutoring tools
- Scientific research applications
That distinction matters — a coding assistant answering factual questions is not regulated; a named persona with a backstory that you chat with daily about your feelings is.
What are Doubao, Qwen and Tencent actually doing?
Three of the largest Chinese AI apps chose compliance-by-removal over filing under the new regime.
- ByteDance's Doubao, China's most-used AI chatbot with about 345 million monthly active users in Q1 2026, is switching off its user-created AI agent and persona feature on 15 July. Users have until 15 October 2026 to export chat data; after that, records tied to deleted agents are non-recoverable.
- Alibaba's Qwen (about 166 million monthly users) started disabling user-created agents on 10 July and finishes the shutdown on 15 July. It is deleting associated chat histories with no grace period.
- Tencent's Yuanbao removed a similar feature in June 2026, ahead of the deadline.
ByteDance has not exited companionship entirely — it is quarantining those features into a separate app called Maoxiang, with strict age-gating. The pattern to watch: not the death of AI companions, but their migration into age-verified, licensed silos.
What compliance obligations does the law impose?
For any operator who does want to keep offering emotionally interactive AI, the checklist is demanding:
- Algorithm filing and security assessment with the CAC before the service can operate.
- Clear AI disclosure: the service must state that the user is talking to a machine, not a person.
- Anti-addiction measures: session-length reminders and cooling-off prompts after prolonged use.
- Crisis-intervention pathways: detection of self-harm signals and a route to human help.
- Protection of minors: virtual romantic or intimate relationships are prohibited for users under 18, and parental consent is required for users under 14.
- Training-data restrictions: private chat data cannot be used to train models without explicit user consent.
The CAC says more than 3,500 AI products have already been removed from Chinese app stores for noncompliance.
How does China's approach compare with the West?
The closest Western analogue is California's Companion Chatbot Law (CCC Law), effective 1 January 2026, which requires AI disclosure to minors and periodic notifications. That is narrower than China's measure: California focuses on notice, while China imposes architecture-level obligations across all users. The EU AI Act treats emotionally manipulative AI as a prohibited practice in narrow cases but does not create a dedicated licensing regime for AI companions.
In short, China is the first jurisdiction to treat "AI that acts like a person" as its own regulated category.
What does this mean for users and for the broader industry?
For roughly 440 million Chinese AI app users, the immediate effect is that a familiar surface goes dark. Custom agents and long-running AI personas disappear from mainstream apps; general assistant features stay. For the industry outside China, the signal is regulatory: governments have started to distinguish between AI-as-tool and AI-as-relationship. Safe-by-default disclosure, session limits, and self-harm routing may become standard requirements. This is the same debate playing out in coverage of AI agent safety and autonomous AI agents — not whether these systems work, but under what constraints they are allowed to.
The Chinese approach also lands against a wider backdrop of AI economic and legal pressure — from warnings about AI job displacement to infrastructure investment like Meta's $50 billion Louisiana data centre and the Apple vs OpenAI trade-secret lawsuit. Regulation is arriving in parallel with the build-out.
Frequently Asked Questions
Q: Is China banning all AI chatbots? A: No. The measures apply only to AI for sustained emotional interaction. Customer service bots, work assistants, Q&A systems, education tools and research applications are explicitly excluded.
Q: When exactly does the China AI companion ban take effect? A: 15 July 2026. The rules were issued on 10 April 2026 by five regulators led by the CAC.
Q: Will my Doubao chat history be deleted? A: Chat data tied to user-created agents becomes non-recoverable after 15 October 2026. Qwen is deleting histories with no grace period.
Q: Can Chinese companies still offer AI companion apps at all? A: Yes, but only under the new regime: algorithm filing, AI disclosure, anti-addiction measures, self-harm intervention, minor restrictions, and training-data consent. ByteDance's separate app Maoxiang is an example of this licensed model.
Q: How is this different from California's Companion Chatbot Law? A: California's CCC Law (effective 1 January 2026) focuses on notification and minors. China goes further, requiring architecture-level changes: algorithm filing, anti-addiction design and training-data consent for all users.
Q: Does this affect users outside China? A: Not directly, but the compliance patterns — mandatory disclosure, session limits, crisis pathways, minor protections — will likely influence EU, UK and US regulators already eyeing emotional AI.
Discussion
0 comments