Steganography on AI Pulse https://cuigh.com/tags/steganography/ Recent content in Steganography on AI Pulse Hugo en-US Wed, 01 Jul 2026 20:47:18 +0800 Anthropic's hidden channel isn't a bug, it's a precedent — and the rest of the industry will copy it https://cuigh.com/posts/claude-code-steganography-china-fingerprint-2026/ Wed, 01 Jul 2026 20:47:18 +0800 https://cuigh.com/posts/claude-code-steganography-china-fingerprint-2026/ Claude Code's client binary embedded a 3-bit steganographic channel in its system prompt for 90 releases over three months, and the v2.1.197 fix shipped with an empty changelog. The technical story is interesting; the governance story is the one that matters. Anthropic is the first major AI lab to walk the line 'client-side binary + silent + export-control = compliance' to completion. Every other lab that wants to identify unauthorized use now has a defensible template. Once a precedent is set, it is much harder to walk back than to set.