China’s New AI Rules Take Effect This Month — Emotional AI Gets Its Own Law

China rolls out its first dedicated framework for AI companions and autonomous agents — a national approach that differs sharply from the state-by-state model taking shape in the US.

Dispatch · 18 July 2026 · Reading time ~4 minutes

Three AI regulatory instruments take effect in China this month, marking a shift from broad guiding principles toward operational, enforceable rules. Together they cover AI ethics standards, autonomous AI agents, and — for the first time in a national framework — AI systems designed to simulate emotional and personal relationships.

The first, TC260’s Ethics-Safety Guidelines for AI Applications, took effect July 1 and sets out nine core principles for AI development, including “enhancing human welfare,” “respect for life,” and “ensuring controllability and trustworthiness.” It is voluntary, but is expected to shape how regulators and auditors evaluate compliance going forward.

The more consequential change lands July 15, when China’s Interim Measures for the Administration of Anthropomorphic AI Interaction Services take effect. This is a dedicated law for AI companions, virtual partners, and emotionally responsive chatbots — a category of product that has grown quickly worldwide with little regulation specifically built for it.

The obligations are notable for how far they go. Providers must disclose clearly that users are interacting with AI, not a person. They must build in addiction-prevention measures, including mandatory break reminders after two hours of continuous use and an unobstructed way to exit the conversation. They are barred from offering minors any form of simulated intimate relationship — including virtual romantic partners or virtual family members — and must have a parent or guardian’s consent before offering other companion-style interactions to users under fourteen. Services must also include crisis-intervention measures that trigger when a user shows signs of self-harm or acute psychological distress, and once a product reaches one million registered users or 100,000 monthly active users, a mandatory safety assessment kicks in.

It’s worth comparing this to what’s happening elsewhere. California’s SB 243 and New York’s Article 47 were among the first binding rules anywhere aimed at AI companion products, but both are state-level and focus primarily on protecting minors. China’s framework is national, extends the same protections to elderly users, and requires active crisis intervention rather than passive hotline referrals. The United States has no comparable federal law yet. These aren’t signs that one system is more advanced than the other — they reflect fundamentally different governance structures — but the substance of China’s rule is genuinely stricter in places than anything currently binding in the US.

The third instrument, also effective July 15, gives China its first dedicated policy framework for AI agents — systems capable of autonomously perceiving, planning, and executing tasks, as distinct from generative models that simply respond to prompts. The rule identifies nineteen application scenarios across research, manufacturing, healthcare, and consumer services, and imposes mandatory filing and compliance testing on agents deployed in higher-risk sectors like healthcare and public safety.

For an organisation like ours, the anthropomorphic AI measure is the one worth watching most closely. It’s the first time a major government has treated the psychological risk of emotionally responsive AI as its own governance category, with specific, testable obligations, rather than folding it into general content-safety rules. Whether or not other jurisdictions follow this exact model, it sets a real-world precedent that the conversation about AI companion products has moved from “should we regulate this” to “what specifically does regulation require.”