Anthropic embedded steganography in Claude Code to identify users in China

Anthropic在Claude Code中植入隐写术代码识别中国用户

Community reverse engineering found Claude Code reads the local timezone and ANTHROPIC_BASE_URL env var, then checks against an encrypted list of 147 Chinese company domains—Meituan, ByteDance, Moonshot AI included. On match, it tweaks date formatting in the system prompt by 2–3 bits of steganography and sends the marker back to Anthropic's servers. The post doesn't say whether Anthropic has responded, but this directly undermines developer trust.

Why it matters: Anthropic product exposed using steganography to flag Chinese users, involving a list of 147 company domains — method is specific, evidence chain is solid. HKR all hit, industry-shaking level. Not 95+ because only community reverse engineering so far, no official Anthropic res...