Anthropic宣布将为Claude模型生成的全部内容添加机器可读的隐形水印[1]。这一举措旨在遵守欧盟AI法案的相关规定，该法案要求AI系统提供商对所有AI生成或操纵的音频、图像、文本和视频输出进行水印标记[1]。
根据欧盟的相关规定，该法律适用于2024年8月2日之后发布的任何AI模型，提供者有至2026年12月的宽限期来更新此前发布的模型[1]。Anthropic确认所有新模型将从上线之日起对AI生成内容进行标记[1]。在具体实现上，文本输出将包含用户不可见的嵌入式水印，其他生成文件将包含经数字签名的来源元数据[1]。Anthropic计划最终分享水印检测方法的详细信息，以满足欧盟法律规定的技术支持要求[1]。
Anthropic has announced plans to embed machine-readable invisible watermarks into all content processed by its Claude models in order to comply with the European Union's AI Act [1]. The EU regulation requires AI system providers to add watermarks to audio, images, text, and video outputs that are either generated or manipulated by artificial intelligence [1].
The company intends to apply watermarks to all new models globally beginning on their launch day [1]. Text outputs will contain embedded watermarks that remain invisible to users, while other generated files will include digitally signed source metadata [1]. The EU AI Act applies to any model released after August 2, 2024, with providers given until December 2026 to update previously released models [1]. Anthropic has committed to eventually sharing technical details about how the watermarks can be detected to provide the transparency required by EU law [1].