LobstersThursday · July 2, 2026FREE

Claude Code Is Steganographically Marking Requests

claudesteganographyprivacyai-tools

A blog post by thereallo.dev reports that Claude Code, Anthropic's command-line AI coding assistant, embeds a steganographic marker in its requests. The marker consists of a string of whitespace characters appended to the end of prompts, which encodes a timestamp and a unique identifier. This hidden data is invisible to users during normal interaction but can be extracted by examining the raw prompt text. The author discovered this by comparing a prompt sent to Claude Code with the same prompt sent via the Anthropic API, noting the extra whitespace. The post suggests this practice may be used for tracking or identification purposes, but the exact intent is not stated in the source. The consequence highlighted is that users are unaware of this hidden data being transmitted, raising questions about transparency and consent in AI tool usage.

// why it matters

Hidden markers in AI tool requests raise transparency concerns for developers relying on these tools.

Sources

Primary · Lobsters
▸ Read original at thereallo.dev

Like this? Get the next digest.