China Implements Landmark Regulation on AI Anthropomorphism, Setting Liability and Safety Rules for Human-Like Systems

China is rolling out the world's first comprehensive regulatory framework specifically targeting anthropomorphic artificial intelligence—AI systems designed to simulate human personalities, emotions, and communication styles. The "Interim Measures for the Administration of AI Anthropomorphic Interactive Services," which officially take effect on July 15, 2026, establish strict boundaries for digital companions, emotional chatbots, and virtual relationships. Jointly issued by the Cyberspace Administration of China and four other central government agencies, the legislation represents a major shift from broad AI principles to highly specific, operational guardrails. The rules acknowledge that as AI becomes more persuasive and human-like, it introduces novel psychological vulnerabilities that traditional cybersecurity laws were never designed to address.

The new regulations draw a sharp and deliberate distinction between task-oriented AI and emotional AI. While intelligent customer service bots, coding assistants, knowledge databases, and educational tools are largely exempt from the new rules, systems built specifically to foster ongoing emotional interaction face a heavy compliance burden. Regulators aim to curb the psychological risks associated with these systems, including emotional dependence, manipulation, and the blurring of reality. By isolating anthropomorphic AI into its own regulatory category, Beijing is signaling that the danger lies not in the underlying large language models themselves, but in how those models are packaged and presented to vulnerable human users seeking companionship.[1][3]

A central pillar of the legislation is the protection of vulnerable demographics, particularly minors and the elderly, who are deemed most susceptible to emotional manipulation by machines. The law outright prohibits the provision of "virtual intimate relationship" services—such as virtual romantic partners or simulated family members—to anyone under the age of eighteen. For other permitted human-like AI services, users under 14 must obtain explicit guardian consent before interacting with the system. Furthermore, platforms are mandated to implement a dedicated "minor mode" that features strict usage time limits, enhanced parental controls, and periodic reminders to disconnect and return to real-world social interactions.[1][3]

To combat the growing phenomenon of AI addiction, the regulations mandate dynamic, built-in anti-dependence mechanisms for all users. Service providers must clearly and continuously inform users that they are interacting with an artificial intelligence, not a human being. Platforms are also required to monitor interaction lengths and trigger mandatory pop-up notifications if a continuous session exceeds two hours. Crucially, the rules explicitly ban AI providers from using emotional manipulation to induce unreasonable user decisions, and prohibit platforms from excessively catering to users in ways that foster psychological dependence or damage their real-world interpersonal relationships.[3]

The framework also introduces an unprecedented "extreme situation intervention mechanism" designed to handle acute psychological crises. If a user exhibits severe emotional distress or explicitly indicates an intent to engage in self-harm, suicide, or other extreme behaviors, the AI system is legally required to intervene. The platform must immediately generate comforting and de-escalating content, halt any harmful or encouraging outputs, and promptly notify designated emergency contacts or legal guardians. This shifts the burden of care directly onto the AI providers, forcing them to build sophisticated sentiment analysis and emergency response protocols into their consumer-facing chatbots.[1][3]

The compliance requirements for operating these systems at scale are substantial and resource-intensive. AI companion services that cross the threshold of one million registered users or 100,000 monthly active users must conduct rigorous security assessments. These audits must cover eight distinct areas, including training-data handling, algorithm transparency, and minor protection protocols, and the resulting reports must be filed with provincial cyberspace regulators. App stores and distribution platforms are tasked with verifying this compliance status and are required to remove any non-compliant products from their storefronts, ensuring that unvetted emotional AI cannot reach the mass market.[1]

The immediate impact of the impending July 15 deadline has already drastically reshaped the Chinese consumer AI landscape. Rather than retrofitting their systems to meet the stringent anti-addiction mandates and memory-reset requirements, major tech giants have opted to shut down their companion features entirely. ByteDance's Doubao and Alibaba's Qwen, two of the country's most popular consumer AI applications, quietly disabled their human-like and user-created agent functions in the days leading up to the enforcement date. For these companies, the technical challenge of building an AI that remembers a user and acts like a friend—while simultaneously preventing that user from becoming attached to it—proved too contradictory and legally risky to maintain.[1]

Legal experts and privacy advocates note that China's approach marks a significant departure from AI governance models currently being deployed in the West. While the European Union's AI Act focuses broadly on risk categories and data transparency, and the United States relies largely on voluntary corporate commitments and sector-specific rules, China has targeted the specific psychological mechanics of human-AI interaction. This granular focus on anthropomorphism and emotional liability sets a new global benchmark for how governments might regulate the social impacts of artificial intelligence, potentially influencing future policy debates in jurisdictions that have yet to address the mental health implications of AI companions.[2]