China's Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services took effect on 15 July — the world's first national regulatory framework specifically targeting AI services that simulate human characteristics and emotional interactions.
The rules target AI services that 'simulate human personality traits, thinking patterns and communication styles to provide sustained emotional interaction.' Customer service bots, knowledge tools, workplace assistants and educational applications are explicitly excluded unless they involve sustained emotional interaction — a distinction that draws a regulatory line between utility AI and companion AI.
The impact was immediate. ByteDance's Doubao informed users that its agent feature would go offline on 15 July due to 'product function adjustments.' The feature had allowed users to customise assistants, tutors, role-playing characters and companions with distinct personas and speaking styles. Alibaba's Qwen issued a similar notice, disabling humanlike interactive agents and user-created agent functions on 10 July, with broader agent services taken offline on 15 July.
The data consequences are permanent. Doubao stated that after 15 October, associated data would be handled per its privacy policy and 'no longer be viewable or recoverable inside the app.' Qwen users lost access to agent settings and previous conversations immediately after shutdown.
The regulatory logic is architectural rather than cosmetic. The Interim Measures require anti-addiction systems, mandatory usage notifications and instant-exit mechanisms that are fundamentally incompatible with how persistent-memory agents are designed. An agent built to maintain a consistent emotional relationship with a user over time cannot simultaneously implement the friction the regulation demands. Rather than retrofit their systems, both companies chose complete feature removal.
The timing is significant. China's AI companion market had been growing rapidly, with millions of users building emotional relationships with customised AI agents. Beijing's intervention reflects concerns about psychological dependency, data privacy in intimate AI interactions, and the broader social implications of AI systems designed to simulate human connection.
The rules arrive as Chinese AI companies simultaneously dominate global open-weight model usage — commanding 61 per cent of tokens on OpenRouter — while facing unprecedented domestic regulatory constraints on how those models can be deployed in consumer products.
For context engineers, China's companion AI rules represent the first major regulatory test of the boundary between AI as tool and AI as relationship. The distinction matters: as agentic AI systems become more capable and persistent, every jurisdiction will eventually need to decide where utility ends and emotional simulation begins.
