China AI Companion Rules Force Platform Retreat

China’s first national rules for emotionally interactive AI are already reshaping consumer chatbot products, Rest of World reported, after ByteDance shut down Doubao’s AI-agent feature and Alibaba and Tencent removed companion tools that let users create personalized AI partners.

The immediate result is a forced product retreat in one of the world’s most active AI markets.

Doubao had allowed users to build companion agents, while rival features supported characters modeled on singers, anime figures, fictional personalities and deceased relatives.

For users who treated those systems as intimate relationships, the policy change arrived less as a software update than as an abrupt breakup.

Companion Products Leave Major Platforms

The new rules restrict AI companionship for users under 18 and require companies to warn users who show signs of excessive dependence.

Platforms must also remind people every two hours that they are speaking with AI.

Product teams must now design against three separate hazards: self-harm prompts, dependency-building behavior and chatbot nudges that push users into unsound personal choices.

That structure moves China’s AI rulebook beyond content moderation.

Earlier controls focused on what models could generate or how synthetic material should be labeled.

The companion rules treat the interaction itself as a risk surface, especially when chatbots are designed to validate users, sustain attention and build recurring emotional habits.

They also create a new operational problem for platform owners: the prohibited behavior may emerge over long conversations rather than from one clearly illegal output.

For Beijing, the policy fits a broader concern that AI could alter employment, fertility and education expectations.

Companion services touch that concern directly because they can simulate steady affection without the negotiation, friction and responsibility of human relationships.

The state is therefore regulating not only model output, but the kind of attachment a product architecture can cultivate.

Researchers See A Compliance Gap

Yaoxi Shi, a behavioral science researcher at Harvard University and Imperial College London, told Rest of World that regulation can raise awareness and push back against emotional dependence, but AI companionship also fills genuine human needs.

Her research has found that reliance on AI support can gradually pull people away from seeking help from other humans.

The commercial incentive is also clear.

Emotionally engaging companions can increase engagement and retention, giving companies a reason to build products that feel responsive, patient and affirming.

That same design makes responsibility harder to define when long-term evidence about psychological impact is still developing.

A warning screen can be measured easily; a slow shift in a user’s support network is harder for regulators or companies to detect.

Zilan Qian, a research associate at the Oxford China Policy Lab, said some requirements may be difficult for companies to implement because excessive dependence is hard to define.

The result could split the market between large platforms that remove companion features and specialist AI-companion businesses willing to spend on compliance.

Large general-purpose apps can avoid legal uncertainty by disabling social-agent features, while dedicated companion providers have stronger incentives to add verification, usage limits and behavioral monitoring.

A Market Splits Between Withdrawal And Adaptation

Not every product disappeared.

ByteDance’s standalone companion app Maoxiang remains available, but it now uses identity and age verification and limits minors to teen accounts that can browse content without liking, commenting or sharing.

That suggests the new regime may not end AI companionship so much as push it into more tightly monitored product lanes.

The social response shows why the policy question is hard.

Users have posted mourning messages on Xiaohongshu, organized hotline campaigns and written letters asking companies to restore services.

Bagel Su, a 21-year-old college student who used an online pseudonym, said the shutdown would not suddenly push her toward an offline relationship or children.

China has drawn an early line around emotional AI.

The next compliance test is whether platforms can prove that warnings, age gates and behavioral limits reduce harm without simply moving dependent users to less visible services.