Anthropic has announced it will embed invisible watermarks in all new Claude models starting in August 2026. The company said the watermarks will be applied globally to both text and files, helping users identify AI-generated content. The move follows Anthropic's signing of the EU AI Act Code of Practice, which requires transparency for AI-generated material. The watermarks will be applied at the model level, ensuring they persist through copying and pasting, though they may be affected by heavy editing or format changes. The company also plans to release verification tools for users and third parties to check the labels, though no timeline has been provided for their release. Source: thedecoder
Anthropic will use two types of labels to identify AI-generated content. Text generated by Claude will carry an invisible watermark that does not alter its meaning, quality, or readability. The watermark remains intact during copying and pasting and may persist through some editing. For supported files like images, the company will apply signed provenance metadata based on the C2PA standard, which can reveal tampering. These labels will be available through cloud partners such as AWS, Google Cloud, and Microsoft Foundry, though support for signed metadata may vary. Source: thedecoder
Anthropic acknowledged the limitations of its watermarking approach. A detected watermark does not confirm that Claude generated the content, as users often use the AI for editing or translating their own text. The absence of a watermark also does not prove the content was not generated by Claude, as older models may not have had the feature or the text could have been heavily modified. The company emphasized that the effectiveness of the watermarks will depend on how well they survive editing, reformatting, and translation. Source: thedecoder
