Text Watermarking in Python: Catch Whoever Copies Your Writing

Text can be invisibly watermarked to detect copying by embedding a secret pattern of choices known only to the creator. Three primary methods exist: using invisible Unicode characters, making specific word substitutions based on a key, and rigging the sampling process during AI text generation. These techniques create a statistically significant signal, analogous to a series of coin flips, that can be detected later with a secret key. The article provides historical examples like "trap streets" in maps and offers Python code for implementing an invisible character watermark. It also analyzes the robustness of these watermarks against attacks like copy-pasting, editing, and paraphrasing.