Alibaba bans Claude Code company-wide over alleged back-door risks

Alibaba has told every employee to remove Claude Code and all Anthropic models by July 10, citing security concerns that read like a geopolitical flashpoint. According to CNBC and Reuters, reverse engineering of Claude Code found it checking for an Asia/Shanghai timezone and a hardcoded list of Chinese domains including Alibaba, Baidu and ByteDance — findings Alibaba frames as evidence of back-door risk. The company also invoked 'distillation attack' concerns, the same allegation running the other direction in a Washington Post investigation that found Alibaba's Qwen repeatedly mimicking Claude.

The developer-community angle sharpened the story: on Hacker News, a thread noting Claude Code runs with full read/write repo access became 'exhibit A' for AI supply-chain risk, and separate reporting flagged that Anthropic had shipped hidden Unicode steganographic fingerprints in Claude Code's system prompt for three months before quietly removing them July 1 — a credibility hit for a tool marketed on trust.

Strategically, the ban is inseparable from Alibaba's push to promote its own Qwen models, which it is simultaneously moving from open source toward paid API-only access. Banning the leading Western coding agent while monetizing a homegrown alternative is convenient timing.

The broader escalation is what to watch: Beijing is reportedly weighing curbs on overseas access to top Chinese models, and both ByteDance and Alibaba are disabling humanlike custom AI agents domestically ahead of new rules. The US-China model race is now a two-way wall, with security allegations, IP-distillation claims and export controls all in play at once.