Anthropic表示将在其AI模型生成的文本中添加水印功能[1]。这一举措旨在符合欧盟《AI法案》透明度代码的相关要求，该规定于8月2日生效[1]。水印技术将应用于Claude、Claude API、Claude Code、Claude Cowork和Claude Tag等产品[1]。
所有8月2日后发布的Anthropic模型将自动配备水印技术[1]。该公司同时计划为现有的旧版本模型扩展水印支持[1]。这些水印作为文本的组成部分，在用户复制粘贴时会随文本保留，即使在某些编辑操作后仍可能存在[1]。
Anthropic has announced that it will embed watermarks into text produced by its artificial intelligence models, including Claude, in response to transparency requirements outlined in the European Union's AI Act.[1] The regulation takes effect on August 2, with all models released on or after that date automatically incorporating the watermarking technology.[1]
The watermarks function as an integral part of the generated text and will persist when content is copied and pasted, though they may be affected by certain editing operations.[1] The watermarking capability will be applied across Anthropic's product suite, encompassing Claude, Claude API, Claude Code, Claude Cowork, and Claude Tag.[1] The company is also planning to extend watermark support to earlier versions of its models.[1]
This move aligns Anthropic with broader industry efforts to mark artificial intelligence-generated content, as other platforms including Suno and Substack have similarly committed to implementing AI content labeling.[1]