AI now watermarks your TEXT and CODE to trace it later as machine generated

AI is entering a new phase where the content it creates may carry a hidden digital watermark. Anthropic says future versions of Claude will use this technology to help identify whether Claude was involved in creating a piece of text: https://www.anthropic.com/news/claude-text-watermark 

The interesting part is that nothing visible is added to the writing. Yet when you copy its subtantial portion to other place it is likely to carry that watermark! There are no strange symbols, hidden characters, or obvious marks. Instead, the system slightly changes how it makes small word choices. These choices create a pattern that readers cannot see, but a detection system with the right key can recognise later.

This could make it easier to check whether content was produced by AI, especially as AI written articles, assignments, and online posts become more common. However, the watermark does not prove that Claude wrote everything. It only shows that Claude was probably involved at some point.

The same idea can apply to code, although it is more limited because programming often requires exact words and commands. Watermarking may work better in flexible parts such as code comments.

You may checkout the European Union's AI Act, especially the new Article 50 https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-50 That is one of the reasons why AI companies like Anthropic are coming up with solutions like this.

Google tried is long ago - see https://deepmind.google/models/synthid/ and many enthusiasts have solutions to remove such markers https://github.com/guillaumemeyer/watermarks-remover too!

This video explains about how such systems work: 

For users, this means AI generated content could become easier to trace later, even when the watermark remains completely invisible to the human eye.