tech
Anthropic wants AI text to be easier to spot. Developers are already building ways around it.
Anthropic's AI watermarks have prompted developers to create tools designed to remove them from text and files.
TL;DR
- Anthropic is embedding imperceptible watermarks in text from supported Claude models to make AI-generated content easier to identify.
- The watermarks are statistical patterns created through word choices and are designed to travel with copied text.
- Developers are rapidly creating tools to remove these AI watermarks, with interest in 'AI watermark remover' surging.
- Concerns exist that watermarks could apply even to minor AI uses like proofreading or summarization.
- The development of watermark removers highlights limitations of watermarking technology, as they can potentially be bypassed.
- The legal landscape surrounding AI watermarks and their removal is a gray area, particularly concerning third-party removal tools.
- Anthropic aims to meet EU AI Act commitments with these watermarks, which could be applied to older models as well.
- The company's usage policy prohibits misrepresenting AI-generated work as human-made.