China tightens rules on AI companion bots over emotional dependency risks

Beijing’s new rules for AI companion bots are an early attempt to draw a bright line between “tools that help” and “bots that bond.” By targeting services that offer sustained emotional interaction, regulators are acknowledging that anthropomorphic agents create a qualitatively different risk surface – one where dependency, grief and manipulation matter as much as misinformation or jailbreaks. Requirements to detect emotional distress, intervene in crises and limit overuse effectively push providers toward building mental-health-style safeguards into commercial chatbots.([en.people.cn](https://en.people.cn/n3/2026/0716/c90000-20478315.html))

For the race to AGI, this highlights a deeper tension. As labs push toward more agentic, empathetic models that can live alongside people as companions, coaches or co-workers, these same capabilities will trigger stricter oversight. China is signalling that the more your system looks like a relationship rather than a tool, the more it will be treated like a quasi-clinical product. That has knock-on effects for training data, logging, escalation pathways and even business models.

Globally, expect this to be a preview. If emotionally persuasive agents become a standard UX for frontier models, Western regulators will face similar questions around addiction, loneliness and undue influence. Labs that design for “healthy attachment” and transparent boundaries now will be better positioned when companion-style AGI crosses from novelty into infrastructure.