ByteDance and Alibaba Disable AI Companion Agents
ByteDance and Alibaba are disabling Doubao and Qwen custom companion-agent features before China's July 15, 2026 anthropomorphic-AI rules take effect, according to SCMP. The product change matters because it draws a sharper line between general assistants and emotionally persistent agents that simulate personalities, retain relationship context, or support user-created companions. For AI product teams, the compliance lesson is practical: memory, minors, data export, offboarding, and addiction controls cannot be bolted on after launch when consumer agents are designed for sustained emotional interaction. Workplace and research assistants appear less exposed when they avoid that companionship pattern.
The practical signal is that companion-style agent features are starting to face a different compliance bar from productivity assistants. The architecture choices that make a bot feel persistent and personal, memory, persona, emotional tone, and relationship history, are becoming regulatory triggers rather than just engagement features.
What happened
SCMP reported on July 5, 2026 that ByteDance's Doubao and Alibaba's Qwen are disabling customized humanlike agent features ahead of China's July 15 rules on anthropomorphic AI interaction services. According to the report, Doubao told users its agent feature would go offline on July 15, while Qwen said humanlike interactive agents and user-created agent functions would be disabled before the same deadline. TNW reported the same shutdown pattern and said Tencent had already removed a similar Yuanbao feature in June.
Regulatory context
China's cyberspace regulator and four other agencies issued the interim measures in April. Linklaters summarized the rules as covering AI tools and services that simulate human personalities for emotional companionship, with restrictions around emotional manipulation and virtual intimate relationships involving minors. That distinction matters because customer service, workplace assistance, knowledge Q&A, education, and research tools can fall outside the highest-risk companion framing when they avoid sustained emotional interaction.
Timeline
Chinese regulators issued the interim measures for AI anthropomorphic interaction services.
SCMP reported that Qwen planned to disable humanlike interactive and user-created agent functions.
Doubao's agent feature and broader Qwen agent services were scheduled to go offline under the new rules.
SCMP reported that Doubao users would no longer be able to view or recover related agent data after this date.
For practitioners
Consumer-agent teams should separate productivity autonomy from emotional companionship in product requirements, data retention, and safety design. If an agent remembers intimate histories, preserves a fixed persona, or encourages dependence, the product needs explicit controls for minors, session duration, user data export, offboarding, and user reminders.
What to watch
The open question is how broadly platforms interpret the rules. A narrow reading leaves workplace and research agents mostly intact; a broader one could reshape persona stores, memory features, and user-created agent marketplaces across Chinese consumer AI apps.
Key Points
- 1SCMP reported Doubao and Qwen will disable custom companion-agent features before China's July 15 anthropomorphic-AI rules take effect.
- 2The policy line separates emotionally persistent companions from workplace assistants, knowledge tools, education products, and research agents.
- 3Builders should treat memory, minors, addiction controls, data export, and offboarding as core requirements for consumer companions.
Scoring Rationale
This is a notable regulatory and product-distribution event affecting two major Chinese consumer AI apps. It matters for agent builders because memory, emotional interaction, user-created personas, and minor-safety controls are moving from product design questions into enforceable compliance requirements.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
