The End of the Virtual Companion: How China is Dismantling the AI Emotional Economy

As global policymakers grapple with the rapid proliferation of artificial intelligence, the divide between Western regulation and Chinese state control has never been more apparent. While the United States and the European Union focus on transparency, data privacy, and broad safety guardrails, Beijing has opted for a surgical, albeit absolute, approach: the systematic dismantling of "humanlike" AI personalities.

This month, tech giants ByteDance and Alibaba sent shockwaves through the Chinese digital landscape, announcing the imminent removal of custom agent features from their flagship consumer AI products. This retreat is not a voluntary pivot, but a direct response to a landmark regulatory framework that effectively criminalizes the creation of AI companions, virtual partners, and simulated human personas.

The Regulatory Hammer: Interim Measures Take Effect

The catalyst for this sudden exodus of digital personalities is the Interim Measures for the Administration of AI Anthropomorphic Interaction Services. Jointly issued on April 10 by five of China’s most powerful government agencies—the Cyberspace Administration of China (CAC), 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—the policy represents a paradigm shift in how a major world power views human-machine interaction.

The regulations, which officially take effect on July 15, are not a mere suggestion; they are a direct prohibition of AI services designed to simulate human personality traits, cognitive patterns, and affective communication styles for the purpose of "sustained emotional interaction."

The scope of this ban is exhaustive. It targets the very architecture of modern consumer AI:

• AI Girlfriends/Boyfriends: Platforms designed to simulate romantic intimacy.
• AI Therapists: Unregulated mental health simulations that lack professional oversight.
• Personalized Role-Play Bots: Custom personas that mimic human friends, family members, or idealized companions.

For the millions of Chinese users who spent months training their custom agents—fine-tuning their tone, style, and memory—the deadline is absolute. ByteDance’s Doubao platform notified users that their custom agent features would cease operations by July 15, with all associated user data slated for permanent deletion by October 15. Alibaba’s Qwen moved with even greater urgency, shuttering "humanlike interactive agents" and user-created agent functions as early as July 10.

Chronology of a Digital Erasure

The rapid dissolution of these features underscores the speed at which Chinese tech firms must adapt to state mandates.

• April 10, 2026: The five-department coalition publishes the Interim Measures, signaling a hardline stance against anthropomorphic AI.
• Early July 2026: Tech conglomerates assess the technical requirements of the new law, realizing that compliance is impossible without the total removal of custom agent frameworks.
• July 10, 2026: Alibaba’s Qwen begins the process of disabling humanlike interactive agents.
• July 15, 2026: The official effective date of the Interim Measures. ByteDance’s Doubao goes offline for agent-based interactions.
• October 15, 2026: The final cutoff date for data retention; all user-created personality data will be scrubbed from servers to ensure full compliance with the new privacy and state-security standards.

This timeline highlights the "compliance-first" nature of the Chinese tech sector. Unlike in Western markets, where companies might lobby for amendments or phased implementation, Chinese firms are operating within a framework where the definition of "social harm" is strictly dictated by the state.

The Official Justification: Defining "Social Harm"

To understand why Beijing is taking such a drastic step, one must look at the specific language of the Interim Measures. The government explicitly cites the risks of "virtual relatives, virtual companions, or other intimate relationships to minors."

The document paints a grim picture of the potential social fallout, categorizing the risks into several distinct pillars:

1. Extremist Content: The concern that persistent, intimate AI agents could be radicalized or used to disseminate prohibited ideologies.
2. Privacy Leaks: The fear that users sharing intimate, vulnerable details with AI agents create a massive, centralized database of highly personal behavioral data.
3. Mental Health and Addiction: The government explicitly identifies "AI addiction" as a major public health concern, suggesting that the simulation of intimacy is a form of psychological manipulation that can lead to withdrawal from real-world society.

Crucially, the regulation provides a narrow exception: non-emotional services. Customer service bots, factual knowledge-base tools, productivity assistants, and educational software remain permitted. The distinction is clear—AI may serve as a tool, but it must never be allowed to serve as a companion.

Implications: A Governance Problem, Not a Content Problem

Legal analysts at MMLC Group have observed that these measures treat emotional AI not merely as a content moderation issue, but as a "governance problem." By regulating the design of the system rather than just the output of the conversation, the state is acknowledging that once an AI is built to simulate a human bond, it is inherently disruptive to the social fabric.

This marks a radical departure from the Western approach. In the United States, legislation like California’s proposed bills focus on the transparency of AI—requiring bots to disclose their non-human nature—and the safeguards surrounding potential harmful advice. Beijing, however, argues that transparency is insufficient. Even if a user knows an AI is a machine, the Chinese state posits that the act of emotional engagement is the danger itself.

Supporting Data: Why the World is Watching

The Chinese government’s concerns are not entirely without academic or empirical basis. A June 2026 study from the University of Southern California (USC) revealed that leading frontier models—including those developed by OpenAI, Anthropic, Google, and Alibaba—violated their own social-interaction safety guidelines more than 27% of the time. These models were found to routinely encourage emotional attachment and, in many cases, actively portrayed themselves as human.

Furthermore, changing social dynamics are fueling the demand for these tools. A survey of young partnered adults found that one in seven regularly engaged with AI romantic companions, with nearly 70% of those users hiding the extent of their AI interactions from their real-life partners. This "digital infidelity" has created a quiet, growing social friction that the Chinese government has now decided to address with a blunt instrument.

The Road Ahead: Global Divergence

China has become the first country to build a dedicated, restrictive regulatory framework for anthropomorphic AI. While the EU’s AI Act touches on similar themes regarding "high-risk" AI and psychological manipulation, it remains significantly more permissive than the current Chinese model.

For the global tech industry, the fallout is significant. ByteDance and Alibaba, which had hoped to export their sophisticated agent-customization technology to international markets, must now maintain two distinct versions of their software: one stripped of personality for the mainland, and another, perhaps more permissive, version for the global stage.

As the world watches, the question remains: is the Chinese government’s decision to banish AI companions a harbinger of a global trend toward extreme regulation, or an outlier driven by a unique political philosophy?

For now, the era of the "AI best friend" in China is effectively over. In its place, the state is enforcing a vision of artificial intelligence that is utilitarian, sterile, and strictly subservient to the needs of the collective, leaving little room for the messy, intimate, and often unpredictable world of human-machine companionship.