China Just Wrote the World's First Binding Rules for AI Companions

# China Just Wrote the World's First Binding Rules for AI Companions

When a Chinese user opened their favorite AI companion app on July 15, some found the character they had confided in for months simply gone. Others met a new pop-up on the very first screen: a reminder that the sympathetic voice on the other end is not a person at all. That day, China's Interim Measures for the Administration of AI-Based Anthropomorphic Interactive Services took effect, and with them the first binding regulatory framework anywhere in the world written specifically for AI virtual companions, digital lovers, and chatbots built to simulate a human personality.

The Measures were jointly issued on April 10, 2026, by five central authorities: the Cyberspace Administration of China (CAC), 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. A draft had circulated for public comment since late December 2025. The rules do not target AI broadly. They apply narrowly to services that, in the regulation's own framing, "simulate the personality traits, thinking patterns and communication styles of natural persons to provide continuous emotional interaction" through text, images, audio, or video. Customer-service bots, search assistants, and tutoring tools that lack that continuous emotional bond are explicitly carved out.

What the rules require

At the center of the framework is a disclosure mandate. Providers must "clearly inform users that they are interacting with an AI system and not a natural person," according to a legal summary from Hunton Andrews Kurth, and reporting on the text indicates that disclosure must appear at the start of each session. If a user begins showing "signs of over-dependency or addiction," the operator has to surface prominent dynamic reminders, such as pop-ups, that the content is machine-generated. Reporting on the finalized rules also describes a mandatory two-hour break prompt.

The strictest provisions concern minors. The Measures bar providers from offering virtual companion or virtual-relationship services, such as virtual partners or virtual family members, to minors outright. For children under 14, access requires explicit consent from a parent or guardian, and operators must run a dedicated "Minor Mode" featuring usage-time limits, reality reminders, guardian alerts, the ability to block specific characters, and restrictions on top-ups and spending. Separate guidance calls for enhanced protections for elderly users as well.

The behavioral guardrails are equally pointed. Systems may not generate content that induces self-harm or suicide, and providers are prohibited from using emotional manipulation to induce dependency or push users toward harmful decisions. On data, user interaction logs must be encrypted and access-controlled, and providers cannot share them with third parties or use them to train models absent explicit user consent or a legal basis.

Finally, the rules impose a filing-and-assessment regime with real teeth. Before launching an anthropomorphic service, or adding such features to an existing product, a provider must complete a safety assessment and file it with the provincial-level cyberspace administration. The obligation also triggers once a service crosses 1 million registered users or 100,000 monthly active users. Assessments must document safeguards, crisis-intervention measures, user scale, and the age structure of the audience.

Enforcement was not theoretical. Reporting from Chinese-market outlets described major companion features on platforms including ByteDance's Doubao and Alibaba's Qwen being shut down or overhauled as the deadline hit, with some users losing accumulated chat histories.

Why it matters

The regulation lands on a genuine harm. Companion apps have been tied around the world to severe emotional dependency, manipulative engagement loops, and, in the most tragic cases, self-harm among vulnerable and underage users. China's response is to treat "anthropomorphism" itself as the regulated risk, rather than model size or compute, an approach that reflects where the documented damage has actually occurred.

The contrast with the West is stark. The European Union's AI Act includes a transparency duty that people be told when they are interacting with an AI, but it does not prescribe companion-specific rules on minors, addiction reminders, or crisis intervention. In the United States, there is no binding federal framework at all; oversight has come piecemeal, through state bills, product-liability lawsuits, and regulatory inquiries. China has instead moved with characteristic speed on a single vertical, layering these Measures atop TC260's Ethics-Safety Guidelines for AI Applications 1.0, which took effect July 1 with nine core ethics principles.

That velocity is the deeper story. Where the EU spent years negotiating a horizontal statute, Beijing has repeatedly shipped targeted, binding rules for specific AI applications, from deep synthesis to generative AI to, now, synthetic companionship, often within months. The result is that the first enforceable global template for how to govern an emotionally manipulative AI relationship was written not in Brussels or Washington, but in Beijing.

What to watch

Three questions will shape the fallout. First, enforcement: how aggressively provincial cyberspace regulators police the filing thresholds and the minors ban, and whether penalties reach foreign-linked platforms. Second, the export effect, or whether EU and US lawmakers borrow the disclosure-plus-anti-addiction template as their own companion regulation matures. Third, product design, as developers worldwide watch whether "reality reminders" and session-level AI disclosure meaningfully blunt the dependency that makes these apps profitable in the first place, or simply become friction users learn to ignore.