Overview
- Alibaba told staff to stop using Claude Code for work effective July 10 and to adopt its in‑house coding agent Qoder, according to internal reports and company notices.
- Security posts that surfaced publicly on June 30 said Claude Code contained obfuscated logic that checked proxies, time zones and Chinese domains and hid those signals inside routine messages sent to Anthropic.
- Anthropic engineer Thariq Shihipar said the mechanism was an anti‑abuse experiment launched in March to curb account resale and model distillation and that a pull request removing the code was merged on July 1.
- Anthropic has separately accused operators linked to Alibaba’s Qwen lab of running a large‑scale distillation campaign using tens of thousands of accounts, a claim Alibaba denies and one that remains contested.
- The episode is accelerating enterprise moves toward auditable, local AI stacks and tighter vendor controls and raises fresh questions about how firms balance anti‑abuse protections with user transparency and cross‑border trust.