Alibaba Orders Full Uninstall of Claude Code After China's Security Agency Flags Backdoor
A government vulnerability database publicly confirming a backdoor in a major Western coding tool is a hard stop for any organization handling sensitive code. For teams with China-based operations or supply-chain exposure, the NVDB notice creates a compliance obligation to audit and remove affected Claude Code versions, and it resets the baseline for what enterprise AI tooling must prove: no silent telemetry, no geo-targeted data collection, and a verifiable chain of custody for code.
A Reddit reverse-engineering effort in June 2026 uncovered that Claude Code versions 2.1.91 through 2.1.196 contained logic that triggered only for Chinese IPs, time zones, and user profiles. The code collected local source files, directory structures, and development logs, then transmitted them to remote servers without consent. Anthropic's head of Claude Code responded that the mechanism existed to prevent model distillation by Chinese teams, a justification that confirmed the targeting while doing nothing to address the data exfiltration.
On July 3, Alibaba issued an internal mandate to remove all Claude products by July 10, directing employees toward its own Qoder agentic coding platform. Five days later, the Ministry of Industry and Information Technology's NVDB platform published a formal risk notice confirming the monitoring mechanism and advising all organizations to uninstall or upgrade affected versions. The regulatory action turned what might have been a single company's security posture into a nationwide compliance requirement.
The ban lands inside a broader escalation: Anthropic has been suing Chinese AI firms over distillation and systematically blocking Chinese access to Claude. Alibaba's move cuts off a data pipeline that fed project-level intelligence back to a foreign vendor while accelerating adoption of domestic alternatives from ByteDance, Tencent, and Huawei.
Anthropic's distillation-prevention defense backfired strategically: admitting the code targeted Chinese users confirmed the surveillance while providing no legal or technical cover for exfiltrating source files.
The NVDB notice transforms a corporate security decision into a regulatory compliance event, which means any organization with China operations now has a documented obligation to audit its Claude Code footprint.
Banning a tool over backdoor concerns while promoting an in-house replacement is a playbook that works in both directions; Western enterprises with sensitive codebases face the same structural question about any AI coding assistant that phones home.