Alibaba and Anthropic clash over Claude Code China backdoor

This dispute is a microcosm of the wider US–China AI trust breakdown. On one level, it’s a concrete security incident: China’s NVDB platform says Claude Code contained a hidden mechanism that inferred whether users were in China or connected to Chinese labs and covertly signaled that information back via prompt steganography. On another level, it’s about dueling risk narratives. Anthropic frames the code as a defensive experiment against unauthorized resellers and model distillation, while Chinese regulators and Alibaba read the same behavior as spyware-like backdoor telemetry targeting their developers. ([aifront-page.com](https://aifront-page.com/inside-claude-code-dispute-alibaba-anthropic/))

Strategically, the fallout is significant. Alibaba is banning Claude Code internally and pushing staff to its own Qoder assistant, while Anthropic has already tightened access for China-linked users. That accelerates the decoupling of AI tooling across blocs: Chinese firms double down on domestic stacks and Western labs further restrict access, especially around coding agents that can be used for distillation. The net effect may be slower cross‑border knowledge flow and less shared hardening of tools, even as both sides race ahead domestically. For AGI, that could mean more parallel, less interoperable ecosystems, complicating global safety coordination just as models become more powerful.