Anthropic will watermark Claude-generated text

Anthropic says Claude will mark AI-generated content with two techniques: invisible watermarks in generated text and signed provenance metadata on supported files. The company describes the move as part of its commitments under the EU AI Act transparency code.

For text, Anthropic says the watermark is woven into the output itself, so it is not visible to readers and should not change meaning, quality or readability. Because the mark is part of the text rather than a separate label, it can travel through copy and paste and may survive some editing.

For supported files such as SVG, PNG and JPG outputs, Claude will attach signed provenance metadata based on the C2PA standard. That metadata can signal that a file was processed by Claude and help detect tampering when the label remains intact.

The useful caveat is that these marks are signals, not proof of authorship. A Claude-edited human document may carry a mark, while heavily rewritten or very short Claude output may not be detectable. Anthropic says it is still preparing detection mechanisms and technical documentation, so developers using Claude will need to watch how the implementation lands in real products.