Anthropic Explains Claude Watermarks: EU AI Act Compliance

Anthropic prepared an FAQ to answer some of the questions they received about watermarks.

In short:

• We are implementing watermarks to comply with the requirements of the EU AI Act. Other major model developers have signed the same Code of Practice and will also be implementing watermarks.

• Our watermarking method has virtually no impact on the quality or content of Claude's responses.

• Readers will not be able to tell watermarked text apart from text without it.

• Nothing is added to the text, and there are no hidden characters.

• Watermarks do not require additional tokens and do not increase cost.

• It is impossible to identify a specific person, organization, or chat from the watermark.