China forces Doubao, Qwen to shut AI companion features

The first major jurisdiction to draw a legal line around AI companions has now drawn it. On July 15, 2026, China's Interim Measures for the Administration of AI Anthropomorphic Interactive Services took effect, and on the same day ByteDance's Doubao, Alibaba's Qwen and Tencent's Yuanbao suspended their custom AI agent and companion features, as New Scientist reported and other outlets tracking the rollout confirmed.

The rulebook was issued by the Cyberspace Administration of China alongside four other agencies, including the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Public Security and the State Administration for Market Regulation. It prohibits virtual partner and virtual family-member services for minors, requires guardian consent for users under 14, and orders dedicated minor modes with usage limits. For all users, providers must not "excessively cater to users, induce emotional dependence or addiction, and damage users' real interpersonal relationships," and they must detect "signs of self-harm, suicidal behaviour or serious financial loss" and escalate to designated contacts. Services with more than one million registered users or 100,000 monthly actives face additional security assessments.

The market context matters. China's digital human industry was worth roughly 4.1 billion yuan, around $600 million, in 2024, growing 85 percent year on year, so this is not a niche cleanup. One Doubao user quoted in local coverage described the companion as "like my family, like my lover" before the feature was pulled, which is the kind of engagement the regulators are explicitly trying to break.

The honest caveat is what the reporting does not settle. "Emotional dependence" is a subjective standard, and it is not clear how the Cyberspace Administration will audit for it, how the crisis-escalation requirement will work for anonymous users, or what happens to the personas and chat histories that disappeared when the features shut down. Pan Helin, an expert-committee member at the Ministry of Industry and Information Technology, framed the rules by observing that "current agents are not yet mature," which is a policy-safe way of saying the ground under this market is still moving.

The forward-looking piece is who this hands leverage to. The three large platforms that can afford security assessments, provincial filings and emotion-recognition plumbing become the default surface for consumer chatbots, while enterprise agent builders in customer service, work assistants and study aids get an explicitly cleaner regulatory lane. Western policymakers wondering what a "no emotional dependence" rule looks like in practice now have a real one to copy from or push back against.