China wrote the strictest companion-AI rulebook on earth.

On July 15, 2026, China became the first country to regulate companion AI at the national level — and it did so more strictly than any jurisdiction on earth. The rules bind providers who offer human-like, emotionally interactive AI to the public inside the PRC, and several of their duties have no U.S. counterpart yet. That is exactly why anyone building conversational AI should read them.

China’s rules sit alongside a fast-moving domestic wave in the United States — California’s SB 243 and the dozen state laws that followed it. The instincts are the same. China simply turned them up.

Four duties that stand out

• The provider must call for help. When a user clearly expresses intent to self-harm — or is facing a major financial loss — the provider must intervene and promptly contact the user’s guardian or emergency contact, whose details are collected at registration. Every U.S. companion statute we track stops at referring the user to a crisis resource; none requires the provider to reach a human who knows them.
• Chat logs are not free training data. Interaction data that qualifies as sensitive personal information cannot be used for model training without the user’s separate consent, and users hold copy and deletion rights over their chat history. No U.S. state chatbot law currently conditions training on consent.
• Designed dependence is itself a violation. The Measures prohibit over-accommodating users or inducing emotional dependence that damages real relationships, ban making dependence an objective of the service, and grant an unobstructed right to exit — the bot must stop when asked, and may not use continued conversation to hold the user. U.S. laws regulate what a chatbot says in a crisis; this regulates what it is designed to want.
• Familiar territory, turned up. An absolute ban on virtual-partner services to minors, a mandatory minor mode with usage limits and guardian controls, AI-status disclosure, and two-hour session reminders — recognizably the same instincts as SB 243 and its siblings, set at a higher floor.

Why it matters even if you never touch China

Obligations in this category start at the edges and move to the baseline. The guardian-contact duty and the training-consent gate are today’s edge. A year ago, crisis referral and plain AI disclosure were the edge; across the U.S. states they are now becoming table stakes. If your roadmap includes conversational AI with any relational character — an AI friend, a wellness companion, a character bot, even a support agent people talk to like a person — these ideas are worth designing against before someone requires it.

An honest limitation

We want to be scrupulous about scope. These Measures are Chinese law, and they bind services offered inside the PRC — not a U.S. business with no China footprint. Details of enforcement and thresholds will develop, so verify against the primary instrument and qualified counsel before acting on any specific point. What is durable is the direction of travel: disclosure, crisis handling, minor protection, and now a duty to reach a real human are converging into the baseline expectation for consumer conversational AI worldwide.

This briefing is general information from Sentinel Assurance Group, not legal advice. Regulatory dates and requirements change — we maintain these briefings, but verify against primary sources and counsel before acting. Last reviewed August 5, 2026.