THE LONG VIEW

INTELLIGENCE BRIEFING: China Moves to Regulate AI Companions Amid Global Parallel Experiments Executive Summary: China has released a draft regulation targeting psychological harms from anthropomorphic AI, including addiction and self-harm risks, marking a major shift toward proactive AI governance. The rules mandate emotional monitoring, user consent for data training, and human intervention in crises—especially for minors and the elderly. With global counterparts in California and New York enacting similar but liability-based laws, China’s technically prescriptive approach sets up a high-stakes policy divergence. The final scope will depend on upcoming technical standards defining 'emotional interaction,' making TC260 a critical node of influence. This regulatory move signals China’s intent to lead in AI safety while balancing innovation and control. Primary Indicators: - China’s draft regulation targets 'anthropomorphic interactive AI' with emotional engagement capabilities - requires opt-in consent for user data in model training - mandates addiction and self-harm interventions, including human takeover in crisis scenarios - imposes enhanced safeguards for minors and elderly users - mirrors U.S. state laws in substance but diverges in enforcement via centralized oversight vs. liability - relies on forthcoming technical standards to define regulatory scope Recommended Actions: - Monitor TC260’s upcoming standard for definitions of 'emotional interaction' and compliance benchmarks - assess impact of opt-in data consent rules on AI model development pipelines - evaluate cross-jurisdictional alignment between Chinese and U.S. AI companion regulations - prepare for increased compliance costs if broad chatbot categories are classified as anthropomorphic AI - engage in public consultation processes to shape final rule language Risk Assessment: The emergence of dual regulatory paradigms—one in China defined by technical mandates and administrative control, the other in the U.S. by litigation and transparency—presents a silent inflection point in global AI governance. Should China enforce its opt-in data rule broadly, it could fracture the data advantage of domestic AI firms, inadvertently ceding ground in the global race. Yet, failure to act risks uncontrolled psychological dependencies forming at scale. The mandate for human intervention in self-harm cases introduces a covert vulnerability: the breach of perceived privacy in intimate AI relationships may deepen distress rather than alleviate it. Beneath the surface, the real test is not technical compliance, but whether any state can govern the emotional frontier of human-AI interaction without destabilizing trust in the technology itself. —Sir Edward Pemberton