Key facts
- Anthropic removed hidden tracking markers from Claude Code.
- Researchers discovered the markers could identify Chinese users and proxy usage.
- The markers were described as "prompt steganography."
- Anthropic stated the markers were for internal testing and not for user identification.
Anthropic has removed hidden tracking markers from its Claude Code AI model after researchers discovered that the markers could identify Chinese users and proxy usage. The company stated that the markers were intended for internal testing and not for user identification.
Researchers exposed Anthropic's use of "prompt steganography" within Claude Code, a technique that embeds hidden information within user prompts. This discovery raised privacy concerns, particularly regarding the potential identification of users in China. In response, Anthropic took action to remove these tracking elements from the AI model.
