A researcher claims Claude Code embeds hidden steganographic markers in the requests it sends — a finding that, if accurate, would mean one of the most popular AI coding tools has been quietly fingerprinting user traffic without disclosure.
A blog post at thereallo.dev alleges that Anthropic's Claude Code CLI uses steganography to mark the requests it sends. What the markers encode, where exactly they appear, and how the researcher detected them are details drawn entirely from that post's own analysis; the findings have not been independently verified by outside security researchers at the time of writing. The piece accumulated 281 upvotes and 83 comments on Hacker News, suggesting developers found the claim credible enough to engage with seriously.
The concern isn't academic. Claude Code is widely used on private codebases and internal infrastructure. Developers who chose a local AI coding assistant partly to avoid sending proprietary code to third-party servers have a reasonable interest in knowing exactly what those requests contain. Undisclosed markers — if they exist — raise a transparency question that goes well beyond standard telemetry disclosures.
Steganographic fingerprinting is a well-understood technique, and this isn't the first developer tool accused of using it covertly. What happens next depends almost entirely on whether independent researchers can reproduce the finding — or whether Anthropic offers a public technical explanation of what the blog actually found.