Claim that Claude Code marks requests with invisible characters

Category: Pulse
Status: discussion
Sources: analysis, discussion
Summary: A blog post that reached the Hacker News front page on 2026-06-30 (205 points) claims Claude Code embeds invisible Unicode characters into the requests it sends, which the author frames as a steganographic fingerprint to detect resale and model distillation. The blog was not reachable from the run environment (HTTP 403), so the exact encoding and what it encodes were not independently verified; invisible-character and variation-selector steganography against frontier models is a documented technique in prior research. Anthropic has not commented.
Comments: HN commenters debated whether the marks would drive account bans or degraded "poisoned" output for flagged usage rather than outright blocking, said the behavior was reportedly visible in an earlier source-map leak, and asked whether the marked system prompt is forwarded to third-party endpoints when ANTHROPIC_BASE_URL points at another provider, which would matter for distillation detection.
Why it matters: Hidden request marking in a widely used coding agent raises trust and privacy questions for teams routing Claude Code through proxies or alternative model backends, though the core technical claim is unverified.