China’s AI Companion Law Takes Effect Today: What Doubao and Qwen Users Lose

Starting today, July 15, 2026, China’s first dedicated rules for AI “companion” apps take effect — and the two biggest players, ByteDance’s Doubao and Alibaba’s Qwen, have already pulled the plug on their humanlike agent features rather than try to comply. I’ve been watching this one develop since April, and what strikes me most isn’t the ban itself — it’s the very specific line Beijing drew between an AI that does your work and an AI that keeps you company.

The Interim Measures for the Administration of AI Anthropomorphic Interactive Services were co-issued back on April 10 by China’s Cyberspace Administration alongside four other agencies, giving companies about three months to prepare. Three months, it turns out, wasn’t enough for anyone to want to try building compliant versions. Both ByteDance and Alibaba just shut the features down.

What Actually Changed Today

Doubao told users its personalized agent function was going offline for “product function adjustments” — corporate language for “we’re not doing this anymore.” Qwen moved even faster: its humanlike, user-created agents stopped working on July 10, and its broader agent services followed five days later, right on today’s deadline.

These weren’t small side features either. Both apps had built genuinely sticky companion-style agents — the kind of assistant people talk to daily, that remembers previous conversations and develops something resembling a personality over time. That’s exactly the category the new rules target.

Why Companions and Not Work Agents

This is the part I think gets lost in the headlines. China isn’t cracking down on AI agents broadly — it still wants companies building coding assistants, scheduling tools, and customer-service bots. What it’s restricting is AI designed to simulate an ongoing emotional relationship. The measures specifically require anti-addiction systems, mandatory usage notifications, real-time detection of “unhealthy dependence,” and an instant-exit mechanism a user can trigger at any time.

There’s also a hard line on minors: no content designed to trigger strong emotional attachment in users under 18, and no AI behavior engineered to feel like it could replace a real relationship. Beijing has been openly worried about parasocial attachment to chatbots for a while, and this is the regulatory answer.

Quick Reference: What’s Changing

The Data Question Nobody’s Answering the Same Way

Here’s where it gets uncomfortable if you were actually using either app daily. Doubao is giving users a window — temporary, read-only access to old agent configurations and chat histories through October 15, 2026, after which the company says the data becomes permanently inaccessible under its normal privacy policy. Qwen hasn’t offered anything comparable. Alibaba has simply confirmed that agent configurations and conversation histories get permanently deleted once the shutdown takes effect, full stop.

If you’ve spent a year building out a personalized agent on either platform, that’s potentially gone with no export tool and no clean way to preserve it. I’d treat any AI companion or agent’s memory as ephemeral going forward, wherever you are in the world — regulation can end a product line overnight.

How This Compares to the Western Approach

What I find genuinely interesting is the contrast in method. The United States has been fighting this out state by state — Colorado’s AI Act is a good example of the patchwork approach, with different rules depending on which state you’re in and Congress still debating a federal framework. The EU leans on broad risk-tiering across all AI use cases. China just picked one specific behavior — simulated emotional relationships — and regulated that use case directly, with a hard national deadline every company has to hit at once.

It echoes the tone from this year’s UN Global Dialogue on AI Governance in Geneva, where scientists told governments plainly that no one can currently guarantee advanced AI systems won’t cause harm at scale. Different governments are reaching for different levers in response — some regulate capability, some regulate specific harms. China went after the harm it was most worried about first.

What This Means If You Use AI Companion Apps

If you’re outside China, nothing changes for you directly today. But the pattern is worth watching, because regulators elsewhere have been circling the exact same issue — attachment, addiction mechanics, and how these apps treat younger users. If a rule like this can force two of the largest tech companies in China to shut down a product line in a single day, it’s not hard to imagine similar pressure landing on companion apps in other markets eventually. I wouldn’t build my daily workflow around any AI companion’s long-term memory right now, anywhere.

FAQ

What is China’s new AI companion law?

It’s the Interim Measures for the Administration of AI Anthropomorphic Interactive Services, a rule set that took effect July 15, 2026, regulating AI services designed to simulate ongoing personal or emotional relationships with users, including required anti-addiction and dependence-detection systems.

Why did Doubao and Qwen shut down their AI agents?

Rather than rebuild their companion-style agent features to meet the new compliance requirements — anti-addiction systems, exit mechanisms, dependence monitoring — both ByteDance and Alibaba chose to shut the features down entirely.

Will my Doubao or Qwen chat history be deleted?

Doubao is offering read-only access to old agent data through October 15, 2026, after which it becomes inaccessible. Qwen has not announced a grace period, and Alibaba says agent configurations and conversation histories will be permanently deleted.

Does this law affect AI agents that aren’t companions?

No. The rules specifically target AI designed to simulate ongoing personal or emotional relationships. Work-focused agents — coding assistants, scheduling tools, customer-service bots — aren’t the target of this particular regulation.