Claude Code Is Steganographically Marking Requests
An investigation found that Anthropic's Claude Code tool embeds steganographic markers in its API requests, allowing the company to detect unauthorized use or reverse engineering. The hidden data, inserted into prompts via seemingly innocuous text patterns, raises privacy and transparency concerns among developers. Anthropic has not officially commented on the technique.
Background
- **Steganography** hides data *within* ordinary content (e.g., tweaking punctuation or whitespace), unlike encryption which is visibly scrambled. The claim is that Claude Code subtly alters its own output so later requests can be fingerprinted.
- **Claude Code** is Anthropic’s command-line AI coding agent, released Mar 2025. It runs in a terminal, edits files, and executes shell commands autonomously.
- The blog author (Loïc) reverse-engineered the tool and found what he argues are steganographic markers embedded in the prompts Claude Code sends back to Anthropic’s API — likely to detect misuse, watermark outputs, or prevent prompt-leaking. Anthropic has not confirmed or denied this publicly.
- This matters because steganography in AI tools raises transparency and trust questions: users cannot easily tell if their interactions are being silently tracked, and the markers could be used to identify individuals or block competing uses (e.g., fine-tuning on Claude’s outputs).