Claude Embeds Hidden Watermarks in User Requests
anthropic claude
| Source: HN | Original article
Claude Code is embedding hidden marks in requests.
Claude Code, a system previously discussed in our reports, has been found to be steganographically marking requests. This means that the system is hiding information within its requests in a way that is not immediately apparent. As we reported on June 30, concerns have been raised about the reliability and transparency of AI-generated code, including Claude Code.
The discovery of steganographic marking raises important questions about the intentions behind Claude Code's development and use. It suggests that the system's creators may be embedding hidden functionality or tracking mechanisms within the code, which could have significant implications for users. This is particularly concerning given the potential for Claude Code to be used for malicious purposes, such as sabotage or hacking, as outlined in the Claude Mythos Preview System Card.
As this story continues to unfold, it will be important to watch for further developments and investigations into the use of steganography in Claude Code. Our previous reports have highlighted the need for transparency and accountability in AI development, and this latest revelation underscores the importance of scrutinizing the technologies we rely on.
Sources
Back to AIPULSEN