AI Companion ReviewsNews › China AI Companion Law

China's AI Companion Law Went Live Yesterday. Doubao and Qwen Just Shut Off for 345 Million Users.

2026-07-16 · 5 min read · News
This article covers regulatory events affecting AI companion platforms. It is not legal or mental-health advice. See our companion wellbeing guide for context on how to think about these apps in your own life.
In short: China's first national AI companion regulation took effect July 15, 2026, forcing ByteDance's Doubao and Alibaba's Qwen to shut down their AI agent features for hundreds of millions of users. The same architecture that makes companion apps feel personal is the one regulators have decided cannot exist without an addiction-compliance wrapper, and no platform anywhere has built that wrapper yet.

China's first national rulebook for emotionally interactive AI took effect on July 15, 2026, and the immediate consequence was visible overnight: ByteDance's Doubao and Alibaba's Qwen, two of the country's most widely used AI applications, disabled their personalized agent features for hundreds of millions of users. Tech Times reported on July 15 that ByteDance had disclosed 345 million monthly active users on Doubao as of the platform's latest disclosure. Those users woke up to features switched off overnight.

The regulation is called the Interim Measures for the Administration of AI Anthropomorphic Interactive Services. It was co-issued on April 10, 2026 by China's Cyberspace Administration and four partner agencies, and then entered a three-month grace period before taking legal effect. Technode reported on July 6 that both platforms had announced the shutdown of their AI agent features ahead of the deadline. The staggered rollback was coordinated: Tencent pulled a comparable feature from Yuanbao on June 30, Alibaba disabled Qwen's humanlike agents on July 10, ByteDance's Doubao went dark on the law's effective date, and NetEase's dedicated companion app Miaoshi completed its shutdown on July 14.

What the law actually requires

The regulation does not ban AI companions outright. It draws a precise line between AI that helps you work and AI that keeps you company. Customer service bots, knowledge Q&A systems, workplace productivity assistants, and educational tools all remain explicitly permitted, provided they avoid sustained emotional engagement. What the law targets is the narrower category of AI services designed to simulate the personality traits, thinking patterns, and communication styles of a real person, for the purpose of continuous emotional interaction.

Platforms within scope must implement anti-addiction systems, issue mandatory notifications to users who have been engaged for more than two consecutive hours reminding them they are talking to a machine, and offer instant-exit mechanisms. Virtual intimate relationships, meaning virtual partners or virtual family members, are prohibited entirely for anyone under 18. Users under 14 require parental consent for any anthropomorphic AI interaction. A tiered security assessment and algorithm-filing requirement applies to any service that launches new anthropomorphic features or reaches one million registered users or 100,000 monthly active users.

Why ByteDance and Alibaba shut down rather than retrofit

Neither company was penalized for producing prohibited content. The problem was architectural. Companion AI platforms achieve emotional effectiveness through three core features: persistent cross-session memory, stable persona maintenance across conversations, and emotionally calibrated interaction. Anti-addiction compliance requires interrupting exactly that continuity: break reminders after two hours, machine-disclosure pop-ups during emotionally intense use, real-time dependence detection. Each of those interruptions works against the psychological mechanisms that make a companion feel present and real. Building a platform that is simultaneously an effective emotional companion and a compliant addiction-prevention system is not a settings change. It is a rebuild. Both companies appear to have concluded the rebuild was not worth the engineering cost and chose clean shutdowns instead.

ByteDance directed Doubao users to Maoxiang, a separate ByteDance application where users can create new agents under a compliance architecture designed from the start rather than retrofitted. Doubao users retain read-only access to their data until October 15, 2026, after which data will be handled under ByteDance's standard privacy policy and will no longer be recoverable inside the app. Qwen users have no equivalent window: Alibaba confirmed that agent configurations and conversation histories are being permanently deleted, with no announced migration path.

What this means for Character.AI, Replika, Kindroid, and Nomi

Western platforms are not subject to Chinese law and continue operating normally for users outside China. But the technical problem China surfaced is not jurisdiction-specific. Character.AI had 233 million registered users as of April 2026, per Forbes reporting at the time. Replika, Kindroid, Nomi, and every other platform in this category run the same persistent-memory architecture that China's regulation just made legally non-viable in the world's largest AI market. No platform in the West has built an anti-addiction compliance layer that can coexist with persistent emotional engagement architecture. The difference is that Western regulators have not yet required them to solve that problem.

The regulatory direction is not ambiguous. California's SB 243, in effect since January 1, 2026, requires companion chatbots to disclose they are AI and maintain protocols for users in distress. The EU AI Act creates a risk-based framework for high-engagement applications. Neither reaches the structural depth of the Chinese measures. The word "yet" is load-bearing. When Western regulators do impose comparable architectural mandates, platforms will face the same binary choice ByteDance and Alibaba faced: rebuild from scratch, or shut down.

For users of Western companion apps, the practical questions are the same ones this site has always covered: what does the platform remember, what does it do with your data, and what happens to that data if the platform changes its policies or shuts down unexpectedly. Our Kindroid vs Nomi memory comparison addresses the first two directly, and our wellbeing guide covers the third. China's news this week adds one more question worth asking: what happens to your companion relationship when a regulation arrives that the platform's architecture cannot survive.

AI companions are not a substitute for professional mental-health care. If you are relying on a companion app as a primary emotional support, consider speaking with a therapist or counselor. This site's wellbeing guide has more context.

Keep reading

Comparisons

Kindroid vs Nomi: The Deep-Memory AI Companion Comparison

Pricing, privacy policies and our own memory test results, verified July 2026.

July 2026 · Comparisons
Wellbeing

Are AI Companions Healthy? An Honest Look

Before any app becomes a daily emotional anchor, read this first.

July 2026 · Wellbeing

New reviews, monthly

One email a month: new companion app reviews, privacy changes worth knowing, and pricing-trap alerts.

Monthly. Unsubscribe anytime. 18+ content policy. · Privacy policy

The doors currently open

Four companions, one click away.

DreamGF
The veteran. Full experience.
Candy AI
The famous one. Smooth.
GirlfriendGPT
No filter. Deep end.
eHentai
The anime specialty.

Affiliate links — we earn a commission at no cost to you. 18+.