China's AI Companion Crackdown: What Authors Need to Know

On July 15, 2026, China's Interim Measures for the Administration of AI Anthropomorphic Interactive Services take effect. The regulation targets AI services that build ongoing emotional relationships with users, and two of the country's biggest AI platforms have already started pulling those features. ByteDance's Doubao is turning off its agent function on July 15, while Alibaba's Qwen will remove humanlike and user-created agents on July 10 and wider agent services five days after that. Tencent's Yuanbao pulled a comparable feature in June.

If you write with AI, this crackdown is worth watching. It signals how seriously governments are now treating AI systems that sustain personal relationships with users, and it has implications for the tools you may already rely on.

What the rules actually cover

The measures, co-issued on April 10, 2026 by China's Cyberspace Administration and four partner agencies, apply to services that simulate human personality traits, thinking patterns, and communication styles to create sustained emotional interaction. Customer service chatbots, knowledge Q&A tools, workplace assistants, and education or research tools are excluded, provided they avoid that kind of ongoing emotional engagement.

Providers are explicitly barred from offering virtual companion or virtual family-member services to minors. They must obtain guardian consent before serving users under 14, build dedicated minor modes with usage-time limits, and put enhanced parental controls in place. The rules also require real-time detection of users in acute distress, with obligations to escalate to guardians or emergency contacts when someone shows signs of self-harm or serious financial loss.

Engineering emotional dependence, addiction, or manipulation to induce unreasonable decisions is explicitly prohibited. Services that cross one million registered users or 100,000 monthly active users must run security assessments across eight areas, from training-data handling to minor protection, and file those reports with provincial regulators. App stores must verify compliance status and remove non-compliant products.

Shanghai's internet regulator said on June 26 it had already removed more than 14,000 non-compliant AI agents for impersonation of official entities, vulgar role-play, and unauthorized collection of personal data.

Why this matters for authors who use AI

You may not run an AI companion service, but the direction of travel is relevant. This is the first dedicated national framework anywhere that draws a detailed line around emotional AI. The rulebook addresses harms that regulators elsewhere have only started naming: teenagers forming attachments to chatbots, companion apps harvesting intimate data, and emotional manipulation by automated systems. China's official case points to the Character.AI lawsuits over psychological harm to minors, FTC investigations into companionship services, and European action against Replika as evidence that the concerns are global.

The same instincts are showing up elsewhere. The EU content labelling rules already in force and the AI Act's August 2026 compliance window place their own disclosure and transparency requirements on AI systems. The FTC and California's SB 243 are moving in similar directions. If you publish AI-assisted work through these platforms, or build reader relationships through tools that may eventually sit in the companion category, the regulatory floor is being set now.

What changed for Doubao and Qwen users

Doubao and Qwen did not break a prohibition. They ran into a design conflict. The measures require companion services to run anti-addiction systems, issue mandatory usage notifications, and offer instant-exit mechanisms alongside real-time detection of unhealthy dependence. Those demands sit awkwardly with agents built to remember a user across sessions and maintain an ongoing relationship, and rather than retrofit, both platforms chose to shut the features down.

The cost landed on users. Many mourned the shutdowns on Weibo, describing the agents as long-standing emotional support and lamenting the lack of an easy way to export chat histories. Doubao is letting people view configurations and conversations in read-only mode until October 15, 2026, before the data is processed under its privacy policy and becomes unrecoverable. Qwen users have been given no comparable grace period; agent data is set for permanent deletion.

What to watch next

The measures leave open a few questions that could matter to writers. There is no technical threshold for what counts as emotional interaction, which is exactly why the platforms pulled entire features instead of trying to carve out exceptions. Liability between platform operators and upstream model providers is not settled when a violation stems from model outputs, and users have no right to carry their data out. Enforcement will flesh out those gaps over time.

Governments watching the experiment will have to decide which parts of China's model they are willing to import. The safety half addresses documented harms that are largely unregulated elsewhere. The control half wraps content restrictions in the same language of user protection, in a package no other regulator would replicate wholesale. Both are real.

If you use AI tools to draft, revise, or publish books, the practical takeaway is simple. Know what category the tools you rely on fall into, read their disclosure and data-export policies, and keep records of how you used them. Transparency is already the standard platforms like Amazon KDP are enforcing, and more regulators are joining that direction. Our guides on how to publish an AI book on Amazon KDP and AI disclosure rules for authors cover the current requirements as they stand.

PageWriter Studio is built for the same reality. Every draft, outline, and chapter you generate stays in a persistent workspace tied to your style profile rather than a generic conversational thread. The tool remembers your project, your voice, and your revision history, and you export the final text on your own terms. If you want to try that workflow on your next title, you can start a free trial and keep control of the record from the first page.