ByteDance, Alibaba Rile Users With End to Human-Like AI Agent Features Amid Regulatory Uncertainty(Yicai) July 6 -- ByteDance and Alibaba Group Holding have both said they will discontinue user-created artificial intelligence companions within their AI apps, citing uncertainty around new Chinese rules governing AI emotional interaction, a move that has spawned widespread dissatisfaction among users.
ByteDance's Doubao and Alibaba's Qwen will remove user-created anthropomorphic AI agent functions on July 15, according to service adjustment notices they released on June 3. These features are distinct from the broader concept of general AI agents and specifically refer to AI personas created by individual users, primarily focusing on emotional companionship.
China’s new Interim Measures for the Management of Anthropomorphic AI Interactive Services take effect on July 15, calling for "inclusive yet prudent and tiered classification-based regulation" of services. But developers have been left uncertain about how to achieve compliance because specific regulatory standards and implementation plans have not yet been released.
Some netizens have criticised the decision by ByteDance and Alibaba, arguing that if the primary concern of regulators is protecting minors, developers could instead require real-name verification or introduce paid access options, rather than shutting down AI companion features altogether.
In response to concerns about data preservation after the features are removed, ByteDance and Alibaba said that users can save their agent configuration information and important conversations through screenshots, chat exports, and other methods. Data would be cleared after a certain period, the pair pointed out.
ByteDance users can transition to Maoxiang, another interactive AI app belonging to the TikTok owner, to create and use AI agent services, it said.
The removal of AI persona agent features does not signal a retrenchment of ByteDance’s and Alibaba’s AI businesses, but is rather a compliance-driven regulatory adjustment, according to analysts. Other firms may also separate high-risk anthropomorphic interaction features from mainstream platforms, and instead operate them through standalone apps better suited to meeting regulatory requirements, they said.
Editors: Dou Shicong, Martin Kadiev
