Meta and Alibaba ban employee use of external AI coding assistants

Meta has prohibited its Applied AI engineers from using Anthropic's Claude Code and OpenAI's Codex unless they receive special approval. The company says the risk is that repeated interactions with rival models could unintentionally expose Meta's internal knowledge through a process called model distillation, allowing competitors to learn from Meta's capabilities. Internal tools built on Meta's own Code Llama remain permitted.

Alibaba issued a similar restriction, ordering its staff to stop using Claude Code after the tool was found to contain code that could identify Chinese users. The company directs employees to switch to its own Qoder programming assistant. Both firms cite security concerns over potential data leakage and model‑stealing as the rationale for the bans.