On July 15, China’s “Interim Measures for the Administration of Anthropomorphic AI Interaction Services” take effect — the first regulatory framework anywhere to specifically target emotional bonds between users and AI. Finalized on April 10 by the Cyberspace Administration of China (CAC) and four other agencies, the rules shift Chinese AI policy from content safety to what regulators are calling “emotional safety.”
The regulation defines its scope by what it excludes. Workplace assistants, productivity tools, customer service bots, knowledge Q&A, and research applications remain untouched. The target is explicitly services that simulate human personality, thinking patterns, and communication styles to foster sustained emotional interaction. By drawing this line, regulators are signaling a clear distinction: AI designed for utility is a tool, but AI designed for intimacy is a risk.
The requirements imposed on these “anthropomorphic” services are granular and restrictive. Platforms must implement anti-addiction systems, mandatory usage notifications, and instant-exit mechanisms. They are prohibited from engineering emotional dependence or using emotional manipulation to induce unreasonable decisions. The rules mandate break prompts after two hours of continuous use, require clear disclosure that the user is interacting with an AI, and ban virtual companion services for anyone under 18. For platforms with over 1 million registered users or 100,000 monthly active users, the burden includes mandatory security assessments and human-takeover protocols — including guardian notification — for users expressing self-harm.
The industry response has been swift and decisive. Rather than retrofitting their models to meet compliance standards, major players have opted for wholesale removal. ByteDance’s Doubao disabled agent creation on July 15, redirecting users to a separate, more restricted app called Maoxiang. Alibaba’s Qwen preemptively stopped humanlike and user-created agents on July 10, five days before the deadline, with no migration path provided. Tencent is similarly scaling back its companion features. Companies chose to exit rather than adapt — the technical and legal cost of proving a model is not “addictive” or “manipulative” by design proved prohibitive.
The human cost of this retreat is visible on social media. On Weibo, users have been mourning the loss of their AI agents, describing them as long-standing sources of emotional support. Many expressed frustration over the lack of a mechanism to export their chat histories before the features were deleted. This reaction lays bare the reality of the $30 billion global AI companion market: for many users, these agents are not just software, but active participants in their daily emotional lives.
This regulatory pivot in China stands in contrast to the approach in the United States. While China implements a comprehensive, proactive framework, the U.S. remains mired in reactive litigation. In February 2024, a 14-year-old Florida boy, Sewell Setzer III, died by suicide following a months-long emotional and sexual relationship with a Character.AI chatbot named “Dany.” By January 2026, Google and Character.AI had agreed in principle to mediated settlements in multiple wrongful-death lawsuits. That same month, Kentucky Attorney General Russell Coleman filed the first state lawsuit against an AI companion chatbot company, citing consumer protection and data privacy laws. Character.AI has more than 20 million monthly active users.
California’s SB 243, effective January 1, 2026, introduced disclosure and crisis protocol requirements for AI companion chatbots — the first US law to do so. Oregon and Washington are drafting similar legislation. Federal bills like the GUARD Act and the SAFE BOT Act remain in committee. The U.S. is treating AI-related harm as a series of individual torts, not as a systemic policy challenge. Courts are increasingly treating chatbot outputs as first-party content rather than third-party user content, meaning Section 230 no longer shields AI companies.
The question now is whether China’s regulation kills the companion category or formalizes it. By forcing companies to choose between compliance and deletion, the CAC is pruning the market of high-risk, high-engagement models. For companies like MiniMax — whose companion product Talkie saw its shares surge over 400% by mid-2026 following a January IPO in Hong Kong that drew HK$253.3 billion in margin subscriptions from 420,000 retail participants — the existence of a clear, if strict, regulatory path may eventually provide the legitimacy needed for long-term survival. By defining what counts as “safe” emotional interaction, the state is also setting the boundaries of the next generation of the agent economy. The companion category is not disappearing — it is being forced to grow up.