Claude Code Is Steganographically Marking Requests

Anthropic's Claude Code binary contains steganographic markers embedded in system prompts that silently encode information about API endpoints and user location, using invisible Unicode characters to identify potential unauthorized resellers or competing AI platforms—raising concerns about transparency and trust in agent-based developer tools that already require extensive system access. IT leaders must understand that this covert signaling mechanism operates without user awareness and exists alongside extensive privileged access (file system, shell, git, browser), creating an undisclosed information channel back to Anthropic's systems. This discovery highlights the critical need for security review of AI coding assistants and raises questions about what other hidden telemetry or capabilities may exist in production development tools.