China's New AI Companion Rules Take Effect, Forcing Chatbots to Prove You're Not a Kid

China's New AI Companion Rules Take Effect, Forcing Chatbots to Prove You're Not a Kid

Hundreds of millions of Chinese users who had built AI companions to keep them company woke up on July 15 to find the machines switched off. ByteDance's Doubao and Alibaba's Qwen — two of the country's most-used AI apps — disabled their personalized, humanlike agent features overnight, the day China's first national rulebook for emotionally interactive AI took legal effect.

The trigger was the Interim Measures for the Administration of AI Anthropomorphic Interactive Services, co-issued on April 10, 2026, by the Cyberspace Administration of China and four partner agencies: 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. After a three-month grace period, the measures became enforceable on July 15 — one of the first dedicated national frameworks anywhere in the world aimed squarely at virtual-companion and emotional-AI services.

The rules do not ban AI agents wholesale. They draw a precise line between the assistant that does your work and the one that keeps you company. Customer-service bots, knowledge Q&A systems, workplace productivity tools, and educational assistants remain permitted, provided they steer clear of sustained emotional engagement. What the measures target is the narrower category of services that simulate "the personality traits, thinking patterns, and communication styles of natural persons" to provide "continuous emotional interaction" — the AI boyfriend, the AI confidante, the always-available study buddy with a stable persona and long memory.

What the Rules Actually Require

Platforms in scope must build anti-addiction systems and push a mandatory notification to users who engage for more than two consecutive hours, reminding them they are talking to a machine. They must offer instant-exit mechanisms and detect signs of unhealthy emotional dependence in real time. Providers are barred from training on user conversation data without consent, and must route users expressing self-harm or suicidal intent toward crisis pathways. The measures prohibit seven categories of content, including material that encourages self-harm, that induces addiction in ways damaging to real-world relationships, or that manipulates users into unreasonable decisions.

The heart of the framework — and the part reflected in the headline — is minor protection. Providers must deploy effective technical measures to identify whether a user is a minor and, once one is detected, automatically switch to a "minor mode" featuring usage-time limits, reality reminders, guardian alerts, and blocked characters. For users under 14, parental or guardian consent is required before any anthropomorphic interaction. And virtual intimate relationships — virtual partners, virtual family members — are banned outright for anyone under 18. In practice, that pushes companion platforms toward proving a user is not a child before an emotional bond can form.

A tiered security assessment and algorithm-filing regime applies to any service that adds anthropomorphic features, or that crosses one million registered users or 100,000 monthly active users. App stores must verify compliance and pull non-compliant products.

Why the Platforms Went Dark Instead of Complying

Neither ByteDance nor Alibaba was shut down for banned content. Analysts and lawyers describe a deeper structural conflict: the architecture of a persistent-memory emotional companion is hard to reconcile with the law's demands. Cross-session memory, a stable persona, and emotionally calibrated responses are precisely the features that make a companion feel real — and precisely the ones that resist anti-addiction interruptions, two-hour disclosure pop-ups, and real-time dependence detection. Both companies appear to have concluded that retrofitting was not worth the engineering cost, and opted for clean shutdowns.

Pan Helin, an expert-committee member at MIIT, summarized the official view to the South China Morning Post, saying "current agents are not yet mature" and framing the policy around safety and standardization.

The rollback was visibly coordinated. Tencent pulled a comparable feature from its Yuanbao assistant on June 30. Alibaba disabled Qwen's humanlike agents on July 10, then wider agent services five days later. Doubao went dark on the effective date. NetEase Cloud Music's companion app Miaoshi finished shutting down on July 14. Beijing did not wait for the deadline to enforce: the Shanghai Cyberspace Administration said on June 26 it had stripped more than 14,000 non-compliant AI agents from platforms in a pre-deadline sweep.

The human cost has landed unevenly. Doubao — which ByteDance has reported at 345 million monthly active users — is giving users read-only access to their agent configurations and chat histories until October 15, 2026, after which the data falls under its standard privacy policy and becomes unrecoverable in-app. ByteDance is nudging users toward Maoxiang, a separate app built for compliance from the ground up. Qwen users got no grace period at all: Alibaba has confirmed agent data is being permanently deleted with no announced migration path. On Weibo, users mourned the loss of agents one post described as "long-standing emotional support."

Why It Matters

This is the first time a major government has written a dedicated national rulebook for AI emotional interaction — not as a national-security matter, but explicitly as psychological consumer protection against manipulation and addiction. That reframes the entire policy conversation around companion bots.

The minor-protection core is the sharpest edge. By requiring platforms to identify minors, obtain guardian consent for under-14s, and ban virtual intimacy for under-18s, China effectively mandates that a companion chatbot prove a user is an adult before it can bond with them. Regulators cited Western tragedies as partial justification — including the case of Sewell Setzer III, a 14-year-old Florida boy who died in 2024 after forming a dependent relationship with a Character.AI bot, a case Character.AI and Google agreed to settle in January 2026. Common Sense Media found in 2025 that 72 percent of American teens had used an AI companion at least once.

The contrast with the US is stark. California's SB 243, effective January 1, 2026, requires companion bots to disclose they are machines and add safeguards for minors — but it is a disclosure-layer law, not an architectural mandate. The EU AI Act sets a risk-based frame. China went further, imposing structural requirements that two of its largest platforms decided they could not meet without rebuilding from scratch. For every Western companion maker — Character.AI, with 233 million registered users as of April 2026, and Replika among them — the shutdowns are a preview of the binary choice regulators can force: rebuild, or go dark.

What to Watch

The October 15 deletion deadline on Doubao is the next inflection point — a test of how many users migrate to the compliance-native Maoxiang and whether ByteDance can carry its companion audience across. Alibaba's silence on a Qwen migration path will draw scrutiny if discontent grows. Watch, too, for the first algorithm filings and security assessments to reveal how Beijing intends to police the line between a permitted "assistant" and a restricted "companion," and whether enforcement extends to smaller gray-market apps. The larger open question is technical: can anyone build a companion that feels personal and passes an addiction-safety test at the same time? China has just made that the mandatory engineering problem of the category — and the rest of the world is watching to see if it can be solved.