Claude’s Text Watermark: How It Works and What It Cannot Prove
Anthropic will add a statistical watermark to text from future Claude models without hidden characters. Here is how detection works, where it weakens, and why a result cannot prove authorship.