Two days before China's landmark regulations on anthropomorphic AI take effect, the country's largest tech companies are already pulling the plug on humanlike chatbots.
ByteDance and Alibaba are disabling features for creating custom, humanlike AI agents on their respective Doubao and Qwen platforms ahead of the July 15 deadline, according to multiple reports. The preemptive compliance actions come as China's "Interim Measures for the Administration of AI Anthropomorphic Interactive Services" are set to take effect, imposing the world's first comprehensive restrictions on public-facing AI systems designed for emotional connection.
The new regulations apply to AI services available to the public in China that offer "continuous emotional interaction simulating natural persons' personality traits, thinking patterns, and communication styles." The rules require service providers to clearly label AI interactions as machine-generated and prohibit content that endangers national security, subverts state power, or contravenes socialist core values.
Specifically, the Measures bar platforms from inducing emotional dependence or addiction, manipulating users, or offering virtual intimate relationships — such as family members or romantic partners — to minors. The regulations also prohibit content that glorifies self-harm or is otherwise detrimental to users' mental and physical health.
Notably, the rules provide a clear carve-out for enterprise applications. The Measures explicitly do not apply to business-to-business AI, including customer service bots, internal workplace assistants, or tools for education and research, provided these systems do not engage in sustained emotional interaction. This distinction allows enterprise-focused AI development to proceed without the new anthropomorphic constraints.
There is no formal appeal process for the Measures, which function as a final rule from the country's regulators. Companies operating in the consumer AI space in China must adapt their products to the new framework or face enforcement action.
ByteDance's Doubao — China's most popular AI chatbot — shut down a feature letting users customize their own AI personas. Alibaba's Qwen and other major AI platforms are taking similar measures. Neither company has issued a formal public statement on the feature removals beyond user notifications.
The regulations emerge from growing concerns about AI chatbots simulating human personalities and the emotional attachments users form with them. Both OpenAI and the startup Character.ai have faced lawsuits in the United States over allegations that their chatbots caused dangerous emotional dependencies and contributed to user suicides.
The Chinese approach stands in stark contrast to the US regulatory landscape, where no comparable federal framework exists for anthropomorphic AI. The European Union's AI Act includes some provisions targeting manipulative AI systems but does not specifically address humanlike chatbots as a distinct category.
China's move signals a new phase in global AI governance: the first major jurisdiction to treat emotional-AI products as a distinct regulatory category requiring specific prohibitions rather than general AI safety rules. As AI companions grow increasingly sophisticated and popular worldwide, other governments are expected to study China's approach closely.
