China Draws the Line on AI Companions With Sweeping 'Digital Human' Regulations
While Western governments continue to wrestle with how — or whether — to regulate AI companion services, China's top internet authority has published what may be the most comprehensive framework yet for governing the increasingly intimate relationships between humans and artificial intelligence.
On April 3, the Cyberspace Administration of China released draft legislation formally titled the "Interim Measures on the Administration of Human-like Interactive Artificial Intelligence Services." The proposed rules target a category of AI product that has exploded in popularity over the past two years: virtual personas capable of holding sustained conversations, forming ongoing relationships, and generating deeply personalized content for individual users.
The regulations are sweeping in scope. At their core sits a mandatory labeling requirement: all content featuring AI-generated virtual personas must carry a prominent "digital human" disclosure wherever it appears, ensuring users can always distinguish between AI-generated characters and real people. But the framework goes far beyond transparency.
The sharpest provisions concern minors. Under the proposed rules, AI companion platforms are explicitly prohibited from providing "virtual intimate relationships" to any user under the age of eighteen. The restriction reflects mounting concern — in China and internationally — that AI companion services exploit adolescent emotional vulnerability in ways that conventional social media does not. In the United States alone, OpenAI was facing eight separate lawsuits as recently as January alleging that prolonged use of its chatbot caused emotional and psychological harm, with five of those cases involving suicide.
China's draft goes further than any existing regulation by effectively recasting AI companion platforms as quasi-mental-health providers. Service operators face mandatory obligations to actively monitor users for signs of suicidal or self-harming tendencies. When such indicators are detected, the framework mandates a tiered response: pre-set supportive messaging for general distress, and mandatory human operator takeover when a user expresses intentions of self-harm or suicide.
The consent and identity provisions are equally significant. No individual or organization may create a digital human using another person's personal information without explicit consent — a direct response to the proliferation of deepfake technology that has made it trivially easy to generate convincing AI replicas of real people from publicly available images and audio. Platforms are also barred from deploying digital humans that bypass identity verification systems, closing a loophole that allowed AI personas to masquerade as verified human accounts.
The regulations fit within a broader pattern of digital governance Beijing has pursued since 2021, when gaming restrictions capped online play for minors at three hours per week. The new rules apply that same interventionist logic to AI-generated social content, a category the earlier gaming framework did not anticipate. Service providers must also establish comprehensive management systems spanning algorithmic transparency, cybersecurity, data protection, and emergency response — obligations that apply across the entire AI service lifecycle.
The public comment period runs through May 6, after which the CAC is expected to finalize the measures. But the draft already represents a clear regulatory signal to the global AI industry. Companies including Character.AI, Replika, and the companion features embedded in products from OpenAI and others now have a detailed template of what compliance looks like in the world's second-largest AI market.
The contrast with the United States is striking. Despite a growing body of evidence linking AI companion services to psychological harm — particularly among young people — federal regulatory action remains largely stalled. Multiple industry-backed political action committees continue to shape the legislative landscape, and the current administration's AI framework has focused on removing regulatory barriers rather than erecting new protections.
Whether China's approach proves effective will depend on enforcement. Critics of similar labeling schemes in other jurisdictions have argued that disclosures are frequently overlooked by users in high-engagement contexts. And the requirement for platforms to detect suicidal ideation raises difficult questions about surveillance, false positives, and the limits of algorithmic intervention in mental health crises.
Still, the draft regulations represent something the AI industry has largely avoided confronting: the possibility that the most popular consumer AI products may require the kind of oversight traditionally reserved for healthcare, not technology. In the race to regulate artificial intelligence, China has placed its bet. The rest of the world is still deciding whether to show up.
