Anthropic 将隐形水印嵌入 Claude 模型层

Anthropic is making invisible watermarks part of Claude at the model level.

Claude models launched in the EU on or after August 2, 2026 will embed machine-readable marks in generated text. For supported models, this applies worldwide across Claude, the API, Claude Code, Cowork, Tag and cloud partners.

Generated files such as PNGs, JPGs and SVGs will also receive signed C2PA provenance metadata where supported.

The text watermark travels with copied content and may survive some editing. Heavy edits, paraphrasing or translation can make it undetectable; short passages may not contain a reliable signal at all.

EU strikes again.