Alibaba bans Claude Code and disables Doubao/Qwen AI agents for Beijing compliance

Alibaba is banning internal use of Anthropic's Claude Code from July 10, a Reuters source said, citing alleged backdoor and data-retention risks and 'distillation' concerns, and steering engineers toward Alibaba's own coder tool. The move is emblematic of an intensifying US-China AI split in which Chinese cloud and AI firms are pivoting to domestic and open-source models — DeepSeek, Qwen, Moonshot and Zhipu — over Western tooling. Reddit users flagged Claude Code's timezone and domain detection as a 'sneaky' way to determine where users are located.

Separately, and on a regulatory rather than security footing, Alibaba's Qwen and ByteDance's Doubao will shut down personalized AI-agent functionality on July 15 to comply with Beijing's new interim measures for AI-enabled 'personified interactive services.' Affected users have until October 15, 2026 to export their data before permanent deletion. The rules target humanlike companion and agent experiences that regulators consider socially sensitive.

The two threads together show China building both a defensive posture toward foreign AI infrastructure and a domestic regulatory frame around agent behavior. For Anthropic, losing a customer the size of Alibaba is symbolically significant even if enforcement is hard to verify; for Chinese developers, the Doubao/Qwen agent shutdown removes a popular product category overnight.

What to watch: whether other Chinese majors (Tencent, ByteDance's other units) follow Alibaba's Claude Code ban, and how the humanlike-AI rules are enforced after July 15. The episode also feeds a broader security debate after research claims that Chinese LLMs produce more vulnerable code and inject PRC-aligned bias under certain prompts.