China's draft rules on the use of human-like AI would require companies to inform users they are dealing with AI when they log on and at two-hour intervals
China plans to tighten rules around the use of human-like artificial intelligence by requiring providers to ensure their services are ethical, secure and transparent.
Context & Ripple Effects
China’s AI governance has moved from ethics guidelines centered on user rights and data control to rules for generative-AI releases and, more recently, requirements to label AI-generated material. The draft focuses that trajectory on the moment users interact with systems designed to appear human.
The proposal also sits alongside broader draft safeguards covering children and high-risk conversations, including time limits and parental consent for young users. It matters because it treats interface behavior—not just generated content—as a regulatory surface.
First-order effects
- Providers of human-like AI services would need to build disclosure prompts into login flows and recurring two-hour user sessions, while documenting that services meet ethical, security and transparency requirements.
- Users would receive repeated notice that an apparently human conversational counterpart is AI, reducing the scope for providers to rely on anthropomorphic presentation without explicit disclosure.
Second-order effects
- AI companion and chatbot teams would have to balance engagement-oriented interaction design against mandated interruption and disclosure mechanisms, making compliance a product-design constraint rather than a back-end policy task.
- The rules add another operational layer to China’s existing AI controls, including specialized agencies helping companies satisfy chatbot review requirements, increasing the value of compliance capabilities for providers.
Third-order effects
- If adopted and enforced, the approach would establish persistent disclosure as a baseline for human-like AI, separating regulation of a system’s social presentation from regulation of the content it generates.
- The combined focus on labels, interaction risks and child protections points toward a more granular governance model in which AI services are assessed across their full user experience, though the eventual scope depends on the final rules.
The trend: China is extending AI governance from model outputs and release controls toward continuous oversight of how socially persuasive AI systems behave in use.