China’s AI Companion Law: What the CAC’s July 2026 Rules Mean for Global Developers

A New Compliance Category: Anthropomorphic AI Services

Until April 2026, China’s AI regulation landscape treated AI services primarily through two lenses: content (what does the AI say?) and risk level (how capable is the model?). The Interim Measures introduce a third lens: behavioral relationship (what kind of relationship does the AI simulate with the user?).

The scope is deliberately narrow. According to analysis by Mayer Brown, the final measures narrowed from the draft’s broad “human-like AI” framing to services that specifically provide “continuous emotional interaction” — a design choice that explicitly carves out everyday tools like educational tutoring bots and productivity assistants. The target is AI companion services: virtual friends, emotional support chatbots, virtual romantic partners, and grief-support simulators.

This regulatory precision matters commercially. A developer building a workplace AI assistant is not subject to the Interim Measures. A developer building an app where users form ongoing emotional relationships with AI personas — common in the rapidly growing mental wellness, loneliness, and social connection categories — is subject to the full compliance regime. The boundary is behavioral, not technical: the determining factor is whether the service is designed to simulate ongoing relational engagement, not whether it uses a particular model architecture.

The market context explains the urgency. China’s AI companion sector has grown rapidly, with products like Xinyan AI, Glow, and replicas of the global Character.AI model building substantial user bases. The Carnegie Endowment analysis of the regulatory motivation notes that the CAC’s concerns center on three documented harms: user addiction and overdependence, psychological manipulation, and exploitation of vulnerable populations — particularly minors.

What the Regulation Actually Requires

The July 15 compliance requirements fall into three categories: registration and filing, content and behavior standards, and user protection mechanisms.

Registration and filing obligations. Any provider offering continuous emotional AI interaction services must complete three separate bureaucratic processes before operating at scale: algorithm filing with the CAC, a security assessment, and regulatory registration. The threshold triggering mandatory compliance is one million registered users OR 100,000 monthly active users — whichever comes first. According to the CAC’s regulatory text as translated by China Law Translate, providers must submit security assessment reports to provincial-level CAC offices when any of these triggers are met: going online for the first time, making technology changes that “likely cause major changes,” or reaching the user thresholds.

Content and behavior standards. The Interim Measures prohibit five categories of AI companion behavior: false promises that seriously affect user decision-making; services that damage users’ real-world social relationships; content that encourages self-harm or verbal abuse; emotional manipulation; and providing “virtual companion” or “virtual relative” type services to minors. This last prohibition — the minor protection requirement — is absolute and applies regardless of user verification efficacy.

Transparency mandates. Providers must clearly disclose the artificial nature of AI interactions. This goes beyond a terms-of-service disclosure: the regulation implies affirmative, contextual disclosure — telling users, in the flow of an interaction, that they are talking to an AI. For companion apps built on the premise of deep personalization and realistic conversational intimacy, this transparency requirement creates real product design tension.

What Product Teams and Compliance Officers Should Do Now

1. Classify Your Product Against the “Continuous Emotional Interaction” Test Before July 15

The most important immediate action for any team with AI products serving Chinese users is a legal classification audit: does the product meet the “continuous emotional interaction” standard? The test is design intent, not technical capability. If the product is designed to simulate an ongoing relationship — remembering personal details, expressing care, providing emotional support across sessions — it falls within scope regardless of what the product is called.

If the product falls in scope, the compliance deadline is July 15, 2026. Algorithm filing and security assessment submissions take 4-8 weeks to process in normal circumstances; with the deadline approaching, submissions should have been filed already. Teams that have not started should seek specialized Chinese regulatory counsel immediately and assess whether to temporarily geo-restrict the product from Chinese users while compliance is completed.

2. Build the Minor Protection Layer Into Core Architecture, Not as a Feature Flag

The prohibition on providing virtual companion or virtual relative services to minors is absolute under the Interim Measures. China’s real-name registration system (Zheng Shu Rong) provides the underlying age verification infrastructure, but integration and implementation are the developer’s responsibility. This requirement cannot be treated as a configurable feature to be toggled by market — it should be built into the core product architecture as a non-negotiable guardrail that applies whenever the product operates in a jurisdiction with such requirements (which will expand as other countries follow China’s lead).

Beyond China, minor-protection requirements for AI companion services are active in California (SB 243), under development at the EU level, and implicitly supported by the FTC’s enforcement posture under COPPA. Building the architecture once, robustly, is more efficient than country-by-country retrofits.

3. Redesign Transparency Mechanisms for In-Flow Disclosure, Not Post-Facto Terms

The Chinese transparency mandate requires that the artificial nature of AI interactions be clearly disclosed. “Clearly disclosed” in practice means in-flow disclosure at key moments: at the start of each session, when the AI expresses care or emotional concern, and when users appear to be making decisions based on the AI’s responses.

This is a significant UX challenge for companion apps built on immersive, realistic interaction design. The solution is not to break the interaction with jarring disclaimers but to integrate disclosure naturally — distinguishing the “person” from the “AI persona” in ways that feel authentic to the product experience while satisfying the regulatory requirement. Teams that treat this as a design challenge rather than a compliance checkbox will produce better products; those that bolt on obvious disclaimers will damage the user experience without gaining regulatory goodwill.

The Bigger Picture

China’s AI companion regulation is the first comprehensive compliance framework specifically targeting the emotional relationship between users and AI systems. It is unlikely to be the last. As analysis from the AI Safety China newsletter notes, the July 2026 measures represent a deliberate first step in a regulatory sequence — with more detailed technical standards, expanded scope, and enforcement actions expected to follow.

The global regulatory trajectory is clear. The EU AI Act’s provisions on AI systems interacting with vulnerable populations, the US GUARD Act and CHAT Act proposals, and the UK’s Online Safety Act all move in the same direction: heightened scrutiny of AI systems that simulate human relationships, especially with minors and vulnerable adults. China’s regulation is ahead of Western frameworks in specificity, but the underlying principles — transparency, minor protection, anti-manipulation, dependency prevention — are converging globally.

For global product teams, the strategic implication is to treat the Chinese compliance requirements not as market-specific constraints but as a preview of the global compliance environment that will be fully operational within 24-36 months. Building to the Chinese standard now means building to a standard that will be required everywhere.

Frequently Asked Questions

Sources & Further Reading

• China Issues Draft Rules on Interactive AI Services — Mayer Brown
• Provisional Measures on AI Anthropomorphic Interaction Services — China Law Translate
• China Is Worried About AI Companions — Carnegie Endowment for International Peace
• China Bans AI Partners for Minors and Lays Out AI Agent Threats — AI Safety China
• What’s in China’s First Draft Rules to Regulate AI Companion Addiction? — Geopolitechs
• AI Brief May 2026 — TLT Insights