China's AI giants curb personalized agents amid regulatory, profit, and security pressures

China's burgeoning artificial intelligence sector faces mounting pressure from evolving regulations, profitability challenges, and U.S.-China tech competition. In response, major Chinese AI firms are scaling back personalized AI services, opting instead for business-to-business solutions and in-house development.

“

The two companies' measures coincide with the implementation of new regulations by Chinese authorities.

ByteDance's Doubao and Alibaba's Tmall, prominent large language model applications, will cease offering "personalized AI agent" features from August 15. These agents, designed to autonomously complete complex tasks like writing and scheduling based on user instructions, are distinct from simple chatbots. Doubao had actively promoted this feature since 2024, while Tmall allowed users to create their own personalized agents.

The timing of this shift coincides with China's implementation of new regulations on "Interim Measures for the Administration of Artificial Intelligence Humanoid Interaction Services" on August 15. These rules target services mimicking human personalities and engaging in emotional exchanges, such as virtual companions or family members. The regulations explicitly forbid content that incites self-harm or suicide, promotes harmful material to minors, or fosters emotional dependency and addiction in users, reflecting a broader effort to control the societal risks posed by AI.

“

The regulations prohibit content that incites self-harm or suicide, promotes harmful material to minors, or fosters emotional dependency and addiction in users.

Beyond regulatory hurdles, financial viability is a significant driver for this strategic pivot. Personalized AI agents, despite high user engagement, incur substantial server and computational costs, making monetization difficult. Alibaba, for instance, has already begun offering its AI agent services to businesses and developers, with clients like KFC and China Eastern Airlines testing the technology. Security concerns also loom large, with Alibaba reportedly instructing employees to remove foreign AI models like Anthropic's Claude due to risks of "backdoor" vulnerabilities, opting instead for its proprietary Qwen service for internal AI tasks.

“

Personalized AI agents are considered 'costly and inefficient' services.

These developments are intertwined with the ongoing U.S.-China AI rivalry. Recent allegations of Claude collecting Chinese user data, though denied by Anthropic as an experimental anti-theft measure, highlight the deep-seated mistrust. Conversely, Anthropic has accused Alibaba and Tmall operators of using their models with numerous fake accounts. The confluence of stringent domestic regulations, economic pressures, and international technological tensions is reshaping the landscape of AI services in China.

“

Alibaba reportedly instructed all employees to delete Anthropic models and agent products due to the risk of 'backdoor' insertion.