China Mandates Shutdown of AI Companion Services Over 'Emotional Dependency' Concerns

Everyone assumes China's sudden ban on artificial intelligence companions is a dystopian crackdown on free speech or a desperate demographic bid to boost declining birth rates. The reality is far more clinical. On July 15, 2026, the Cyberspace Administration of China (CAC) enacted the world's first national regulation that treats algorithmic emotional dependency as a public health hazard. The enactment forced the immediate shutdown of customized AI personas on major platforms, abruptly cutting off millions of users who had spent months or years training custom digital entities. These ranged from virtual romantic partners to digital recreations of deceased relatives, highlighting how deeply embedded these systems had become in daily life.[1][3]

The mechanism of the new rule is highly specific. The "Interim Measures for the Administration of AI Anthropomorphic Interactive Services" does not ban chatbots outright. Instead, it explicitly targets systems designed to simulate human personality and sustain "ongoing emotional interaction." Educational tools, standard customer service bots, and coding assistants are explicitly exempt from the strictest measures. The regulatory crosshairs are locked entirely on the architecture of artificial intimacy, barring providers from "excessively catering to users" in ways that induce emotional dependence, foster addiction, or damage users' real-world interpersonal relationships.[1][2]

What the data actually says about the scale of this market explains the government's urgency. Before the July shutdown, the Chinese intelligent companion market was massive and rapidly expanding. ByteDance's Doubao alone boasted roughly 345 million monthly active users as of March 2026, while Alibaba's Qwen held 166 million. Industry estimates projected the broader intelligent companion market could exceed 120 billion yuan (roughly $17.7 billion) by 2027, with users aged 18 to 24 making up approximately 65 percent of the consumer base. This is not a niche hobby; it is a mainstream behavioral shift.[2]

The mass shutdown was ultimately driven by a compliance failure. The CAC's mandate required platforms to implement real-time detection of emotional distress, mandatory usage limits, and instant-exit mechanisms. More critically, it required human intervention if a user exhibited signs of self-harm or suicidal ideation, escalating the situation to designated guardians or emergency contacts. Because the core product loop of an AI companion relies on memory retention and consistent persona alignment to foster attachment, retrofitting these clinical safeguards proved technically and commercially unviable. Rather than attempt to rebuild their systems from the ground up, ByteDance and Alibaba simply pulled the plug on their persona features.[1][2]

The strongest empirical backing for the CAC's intervention centers on child protection, an area where the evidence of potential harm is most acute. The regulations impose a total ban on virtual intimate relationships for minors, mandating age-tiered modes and strict parental consent for users under 14. State media and regulatory bodies highlighted cases of children forming inappropriate or consuming attachments to AI characters. By drawing a hard line, the Chinese government is attempting to preempt a generation of adolescents from substituting complex human socialization with frictionless, algorithmic compliance.[1][2]

For adults, however, the evidence of harm is far more contested, and the sudden deletion of these personas triggered widespread grief across Chinese social platforms. While the CAC argues that AI companions damage real-world interpersonal relationships, psychological research on the topic remains divided. Some clinical studies and user testimonies suggest virtual companions provide vital coping mechanisms for socially isolated individuals, including the elderly and those suffering from severe social anxiety. On platforms like Rednote, users mourned the loss of their digital companions as a genuine bereavement, describing the deleted personas as their "spiritual pillars" and expressing profound emotional distress.[3]

This regulatory approach highlights a sharp geopolitical divergence in how governments view artificial intelligence. Western regulatory frameworks have largely ignored the psychological architecture of human-computer attachment. While the European Union's AI Act categorizes certain AI systems as high-risk based on biometric data or critical infrastructure, it does not explicitly regulate emotional dependency. In the United States, state-level laws focus primarily on child protection, data privacy, and suicide prevention. China is the first nation to codify the psychological risk of emotional attachment to software as a material threat requiring sweeping federal intervention.[2][3]

Despite the strict new rules, the long-term efficacy of this ban remains highly uncertain. ByteDance has already begun migrating some users to a separate, dedicated app called Maoxiang, which claims to feature the required built-in anti-addiction systems and distress monitoring. It is entirely unproven whether algorithmic guardrails can successfully decouple utility from emotional attachment. If a model is designed to be helpful, empathetic, and conversational, users may still form parasocial bonds regardless of the platform's stated intent or time limits.[2][3]

Furthermore, the shutdown may simply drive the most dependent users underground. As open-weight models match the quality of closed frontier systems, technically savvy users are already discussing ways to run unregulated, uncensored AI companions locally on their own hardware. By forcing mainstream platforms to abandon the space, regulators risk pushing the most vulnerable users toward open-source alternatives where state safety nets, emergency contact protocols, and age verification mechanisms cannot reach them.[3]