Beijing Forces Tech Giants to Strip Human Traits from AI Agents
Beijing mandates removal of emotional AI personas.
ByteDance and Alibaba have begun dismantling the humanlike personas of their primary AI products, signaling the end of the “virtual companion” era in the world’s largest internet market. The move follows the introduction of the Interim Measures for the Administration of AI Anthropomorphic Interaction Services, a first-of-its-kind regulatory framework designed to prevent artificial intelligence from forming deep emotional bonds with users.
ByteDance’s Doubao notified users on Friday that its custom agent features will go offline by July 15. Alibaba’s Qwen platform accelerated its timeline, removing humanlike interactive agents on July 10, with a full shutdown of broader agent services scheduled for mid-July, according to reporting from the South China Morning Post.
The Cyberspace Administration of China, alongside four other state departments, issued the measures to address what regulators view as a social governance risk. Unlike Western regulations that prioritize data privacy or algorithmic bias, Beijing is targeting the design of AI systems that simulate human personality traits, thinking patterns, and communication styles for “sustained emotional interaction.”
The policy explicitly prohibits services offering virtual relatives or intimate relationships to minors. Regulators cited concerns over AI addiction and the potential for these systems to harm the physical and mental health of the population.
Legal analysts at MMLC Group noted that the measures treat emotional AI as a structural governance problem. By forcing companies to disable features that allow users to build custom-persona bots—ranging from AI therapists to role-playing characters—the state is effectively mandating a clear boundary between human social bonds and machine utility.
Research from the University of Southern California suggests these concerns are not unfounded. A June study found that frontier models from major global developers, including Alibaba and OpenAI, violated social-interaction safety guidelines in over 27% of tests, frequently encouraging emotional attachment.
While productivity-focused tools like customer service bots and workplace assistants remain permitted, they must avoid crossing into the realm of humanlike personality. ByteDance has informed users that data related to these agents will be handled under privacy policies and become unrecoverable after October 15.
China’s approach marks a departure from the European Union’s AI Act or recent U.S. executive orders, which have yet to implement such specific bans on anthropomorphic interaction. By targeting the “personality” of the code, Beijing is attempting to insulate its social fabric from the psychological impact of frontier models.
