Show HN: AI Watermark Scrambler, blur statistical LLM text watermarks locally (github.com)

A new tool, the AI Watermark Scrambler, has been introduced to help users locally scramble and blur statistical text watermarks for various large language models (LLMs) like Claude and OpenAI's GPT. Designed to improve the privacy and usability of AI-generated content, this agent skill can alter the word choice in text to reduce the visibility of embedded watermarks—those signals that indicate AI authorship. The tool distinguishes between statistical watermarks and invisible Unicode metadata, enabling users to sanitize their documents while retaining essential content integrity. This development is significant for the AI/ML community as it raises questions about the reliability of watermarking systems meant to maintain the integrity of AI-generated content. The ability to adjust word choices based on green and red lists—administering logit boosts to specific tokens—provides a means to obscure watermarks without completely removing the original text's meaning. However, it should be noted that while the statistical signals can be reduced, they cannot be completely guaranteed to be undetectable, thus underscoring ongoing challenges in watermarking effectiveness and AI content traceability.