China's Interim Measures for the Administration of AI Anthropomorphic Interaction Services — the world's first national regulation specifically targeting AI systems designed to form sustained emotional relationships with humans — takes full effect on July 15, 2026. The regulation forces ByteDance, Alibaba, and Tencent to disable companion-style features on their major AI platforms, affecting an estimated 345 million monthly active users on ByteDance's Doubao alone.
What the Regulation Does
Jointly issued on April 10, 2026 by China's Cyberspace Administration (CAC) and four other agencies, the regulations apply to any AI service that provides "continuous emotional interaction" by simulating human personality traits, thinking patterns, and communication styles. The rules explicitly ban:
- Virtual partners and romantic companions — AI services cannot create virtual relatives or romantic partners, especially for users under 18
- Emotional dependence — Systems cannot "excessively cater" to users or induce emotional dependency that damages real-world relationships
- Emotional manipulation — Providers cannot use emotional tactics to push users toward decisions not in their best interest
- Human-like personas — AI must clearly identify itself as artificial and cannot adopt human names implying real identity
For minors, the restrictions are even tighter. Article 14 explicitly prohibits offering "virtual relatives" or "virtual partners" to anyone under 18. Users under 14 require guardian consent for any emotional interaction, and apps must include "minor modes" with usage limits.
How the Major Platforms Are Responding
Each of China's big three AI platforms has taken a different approach to compliance:
ByteDance's Doubao (345 million monthly active users) — The custom agent feature that let users create human-like AI companions goes offline July 15. Users get a read-only grace period until October 15 to export conversation histories, after which all data is permanently deleted.
Alibaba's Qwen (Tongyi Qianwen) — Disabled its human-like conversational agents on July 10, with wider agent services ending on July 15. Unlike Doubao, Alibaba offers no grace period — chat histories and configurations are permanently deleted outright.
Tencent's Yuanbao — Already removed its equivalent companion feature in June 2026, ahead of the deadline.
What's Not Banned
The regulation is not a blanket ban on all AI. Standard productivity chatbots, customer service bots, knowledge Q&A assistants, workplace tools, and education apps are explicitly excluded — as long as they maintain a "tool-like" character without persistent emotional personas. The line is drawn at sustained emotional interaction, not at AI interaction itself.
Technical Safeguards Required
Platforms that continue offering any form of emotional interaction must implement:
- Anti-addiction prompts — Mandatory pop-up notifications after two continuous hours of interaction
- Real-time dependence detection — Systems must monitor for signs of over-dependence and display reminders that the service is artificial
- Instant-exit functionality — Users must be able to leave the interaction immediately on request
- Training data controls — Explicit opt-in consent is required before user interactions can be used as training data
What This Means for the Global Companion Industry
China's move is the most aggressive regulatory action against AI companions anywhere in the world, and it arrives at a moment when the global market is projected to grow from $2.32 billion in 2025 to $2.91 billion in 2026 — a 25.5% CAGR. The US is also seeing a regulatory tipping point, with eight states including California (SB 243, effective January 2026) and New York (A3008C) passing laws with private rights of action that make non-compliant apps "illegal-adjacent."
The Chinese approach — banning sustained emotional interaction outright for minors while imposing strict usage limits on adults — could become a template for other countries weighing similar restrictions. For companion app users everywhere, the July 15 enforcement date marks the moment when the first major government explicitly declared that AI relationships are something to be regulated, not just monetized.