China Forces ByteDance and Alibaba to Shut Down Humanlike AI Agents

ByteDance and Alibaba are disabling features in their chatbots that let users create personalized AI agents with a fixed personality, name, and conversation style. The companies are acting a week before Chinese regulations governing humanlike interactions with artificial intelligence take effect.

Both apps offered users a pool of agents, created either by the companies themselves or by the community, which could be tailored to specific tasks, skills, and speaking styles. Users could also build their own personas, turning a generic chatbot into a named assistant, tutor, roleplay character, or digital companion with a consistent character and tone.

What exactly is being removed

Doubao told users its agent feature would go offline on July 15, offering only a vague explanation of a "product feature adjustment." Qwen went further, specifying in its announcement that "humanlike interactive agents and user-created agent features" would disappear first, on July 10, with all remaining agent features and services shutting down five days later.

New law on AI anthropomorphization

The timing of both shutdowns is no coincidence. It coincides with the entry into force of China's "Interim Measures for the Administration of Artificial Intelligence Anthropomorphic Interaction Services," announced in April 2026. The rules cover AI services that simulate human personality traits, thought patterns, and communication styles in order to sustain emotional interaction with users.

The regulator explicitly exempts customer service bots, knowledge Q&A tools, office assistants, and educational and scientific tools from the rules, as long as they don't build a lasting emotional relationship with the user. That distinction matters: Beijing isn't banning AI as such, only a specific type of product, the digital companion with a humanlike persona.

The regulator's rationale

Chinese authorities justify the new rules by citing the risk of spreading extremist ideas through personalized agents, privacy leaks from long, emotional conversations, harm to users' mental and physical health, and the risk of dependency on relationships with AI. That reasoning echoes arguments recently raised by regulators and researchers in the West regarding companion chatbots and their effect on young users.

For Doubao and Qwen, this marks the second major regulatory clash in a short span of time. Both companies rank among the top players in China's language model market and compete not only with each other but also with Western players for global enterprise and developer users.

What it means for the market

The decision shows how differently China and the West approach AI regulation. While the European Union's AI Act focuses on systemic risk, transparency, and high-risk classification, Beijing is targeting a specific type of interaction head-on, emotional bonds with a bot, and ordering companies to shut the feature down within days of announcing the deadline.

For Polish companies and developers building products with a conversational layer, this is a signal that the issue of AI anthropomorphization and users' emotional attachment will sooner or later land on the table of European regulators too. The AI companion market is growing very fast globally, and China is among the first major economies to put hard legal guardrails around it.

Sources: South China Morning Post (scmp.com), Decrypt (decrypt.co)