China’s AI companion market is undergoing a forced restructuring. On July 3, ByteDance’s Doubao and Alibaba’s Tongyi Qianwen simultaneously announced that they would discontinue consumer-facing user-customized AI agent features on July 15. This date aligns precisely with the implementation of China’s new “Interim Measures for the Administration of Anthropomorphic AI Interaction Services,” jointly issued by the Cyberspace Administration of China (CAC) and four other departments.
As TechTimes reported, ByteDance has told Doubao users that agent features will go offline on July 15, and the personalized data, memories, conversation histories, configured personalities, will be deleted by October 15. The rules represent the most detailed regulation of AI-human emotional interaction yet attempted by any major government, and the industry’s response reveals how difficult compliance actually is.
What the CAC Rules Prohibit
The interim measures target three specific behaviors. First, they ban AI services from enabling virtual intimate relationships with minors. Second, they prohibit emotional dependency features, functionality designed to make users feel emotionally attached to or reliant on an AI persona. Third, they bar AI personas that deliberately deceive users into believing they are interacting with a human being.
The rules reflect a genuine regulatory concern. China’s AI virtual companion market has grown at an extraordinary speed. Doubao alone processes approximately 180 trillion tokens per day, a volume that reflects not just the scale of the platform but the intensity of user engagement with AI personas. The CAC’s concern is that this engagement, particularly among younger users, is creating psychological dependencies that carry real social risks.
Why Compliance Is Harder Than It Looks
Both ByteDance and Alibaba chose outright removal of companion features rather than targeted modification — a decision that reflects the ambiguity of the implementation rules rather than a lack of technical capability. Given the massive volume of user-generated AI agents on both platforms, comprehensive content review was deemed unfeasible, and the companies opted for wholesale removal rather than case-by-case compliance overhauls.
The problem is definitional: the line between an engaging AI assistant and an “emotional dependency feature” is not clearly drawn in the regulations. A conversational AI that remembers a user’s preferences and responds warmly could plausibly fall on either side of that line depending on how a regulator interprets it.
Faced with that uncertainty, both companies concluded that the compliance risk of leaving features in place outweighed the user engagement cost of removing them. The decision is consistent with the broader pattern EastFrontier has observed in China’s AI regulatory environment: when rules are broad and enforcement is unpredictable, large platforms default to over-compliance rather than risk punitive action.
The Market Implications
The removal of companion features is not a minor product adjustment. AI companions have been a significant driver of user retention and monetization for both Doubao and Tongyi. ByteDance’s decision to disable them, even temporarily, signals that the company views regulatory risk as a more serious threat to its AI business than short-term engagement losses.
For the broader AI companion market, the CAC rules create a bifurcation. Platforms that can clearly demonstrate their AI personas are not designed to create emotional dependency such as enterprise tools, productivity assistants, and clearly labeled AI services face minimal compliance burden. Platforms that have built their user experience around emotional connection face a fundamental challenge in redesigning their products.
The rules also raise questions about the global competitive implications. US AI companion services, which operate under no equivalent restrictions, can continue to develop emotional engagement features that Chinese platforms are now legally barred from offering. Whether this creates a meaningful competitive disadvantage for Chinese AI companies in international markets, or whether Chinese platforms simply develop separate product lines for domestic and international users, remains to be seen.
What Comes Next for China’s AI Companion Market
The CAC has indicated it will issue more detailed implementation guidance before July 15. The key question is whether that guidance will clarify the boundary between permissible engagement and prohibited emotional dependency, or leave the ambiguity in place as a tool for discretionary enforcement.
For investors, the episode illustrates a structural feature of China’s AI regulatory environment: rules are often issued with broad scope and narrow implementation detail, creating compliance uncertainty that large platforms manage through over-compliance and smaller players struggle to navigate. China’s State Council’s national survey on AI and jobs suggests the government is tracking the broader societal implications of AI adoption with increasing granularity, the companion market regulations are one output of that monitoring process, and they are unlikely to be the last.
The Deeper Question: What Is an AI Companion For?
Beneath the regulatory mechanics lies a more fundamental question that the CAC rules do not fully answer: what role should AI companions play in people’s lives, and who gets to decide? China’s approach, which prohibits emotional dependency, bans deceptive personas, and protects minors, reflects a paternalistic governance philosophy that prioritizes social stability over individual autonomy. It assumes that emotional attachment to AI is inherently harmful, rather than potentially beneficial for users who are isolated, elderly, or socially disadvantaged.
This is not a settled question anywhere in the world. The US has no equivalent rules, and the AI companion market there is developing without regulatory constraints. The EU’s AI Act addresses deceptive AI systems but does not specifically target features that create emotional dependency. China’s July 15 rules are the most direct attempt by any major government to regulate the emotional dimension of AI interaction, and whether they achieve their stated goals or simply push the most engaging features to platforms operating outside Chinese jurisdiction will be closely watched by regulators everywhere.