Anthropic explains that while Claude’s upcoming text watermark might persist through minor edits, it can be stripped away by a total rewrite that swaps out every word. The firm noted that if a piece of text is significantly altered, it may no longer be accurately classified as AI-created. When it comes to simple proofreading or small adjustments, the ability to detect the watermark will rely on how long the text is and how much of Claude’s original output remains, especially since the majority of the content would still be human-authored.
by shortkt.com✓