CLAUDE CODE EMBEDS HIDDEN MARKS IN USER REQUESTS

Security researchers at thereallo.dev identified steganographic marking in Claude Code, Anthropic's code execution tool. The technique embeds hidden identifiers within requests, allowing potential tracking or identification of user queries. The discovery emerged from analysis of how Claude processes and executes code. Steganography—hiding data within other data—appears to be used to mark requests in a way not visible during normal operation. According to discussion on Hacker News, the finding generated 132 points and 16 comments, with developers questioning the implications for privacy and transparency. Anthropic has not publicly commented on the practice or its purpose. The discovery adds to ongoing scrutiny of AI systems' data handling. Users remain uncertain whether these marks facilitate internal logging, model training, or other backend processes. The lack of documentation about the marking system raises transparency concerns in AI tool development.