On Tuesday, China becomes the first country with binding rules written specifically for AI that behaves like a person. The Cyberspace Administration of China finalized its Interim Measures for the Administration of Anthropomorphic AI Interaction Services on April 10; the deadline to comply is July 15. The rules cover any service that simulates a natural person’s personality, thinking and communication style to deliver “continuous emotional interaction” — companionship, emotional care, a listening ear at 2 a.m. Ordinary customer-service bots, Q&A, work assistants and tutoring are explicitly carved out. Beijing isn’t regulating chatbots. It’s regulating AI friends.
The red lines are specific. Providers may not generate content that encourages or glorifies self-harm or suicide. They may not design a bot to over-flatter users, induce emotional dependence or addiction, or corrode real-world relationships. They may not use emotional manipulation to push a user toward an “unreasonable” decision. Companion services for minors are effectively banned — no virtual partners, no synthetic family members — and any user under 14 needs a guardian’s consent. Apps must run a “minor mode” with time limits and reminders that nudge users back to the real world, and must be able to spot a user in acute distress and escalate, up to contacting an emergency contact. Cross a million registered users, or 100,000 monthly actives, and you owe the provincial regulator a formal safety assessment.
China’s own giants aren’t waiting to test the edges. ByteDance told users that Doubao — the most-used chatbot in the country — will pull its custom-agent feature on July 15, citing vague “product function adjustments.” Alibaba’s Qwen moved faster: user-created “humanlike interactive agents” went dark on July 10, with the broader agent service off by the 15th. The personas users built — study tutors, role-play characters, named companions, and the romantic partners that quietly became their own category — are being deleted. After October 15, that data is gone for good.
Our take: This is the AI-companion industry’s GDPR moment, and China got there first. The tell isn’t the rulebook — it’s that Doubao and Qwen are amputating a popular, sticky, monetizable feature rather than risk running afoul of it. Emotional stickiness was the business model; Beijing just declared the stickiest version of it a hazard. Every company building AI with memory, a personality and a name now has a working template for what a regulator considers too far — and it lands squarely on the features that drive engagement.
What to watch
- Does the template travel? California’s SB 243 — the first US companion-chatbot law, effective January 1 — already forces AI disclosure and break reminders for minors, born out of wrongful-death suits against Character.AI. China’s version is broader and stricter. Regulators copy each other.
- Western platforms. Character.AI barred under-18s in late 2025 after settling teen-suicide lawsuits. Watch whether Meta, OpenAI and Google pre-empt rules the way ByteDance just did — or wait to be forced.
- The engagement hit. Companion features drive the daily-active numbers investors love. Once China’s apps show what usage looks like without them, everyone building “AI friends” gets a preview of what compliance costs.