Anthropic为Claude全系嵌入隐形水印 覆盖API及云客户
Claude 全系嵌入隐形文本水印，为 AI 生成内容提供官方溯源标记，应对欧盟监管并配套检测工具。
Anthropic 已在 Claude 新模型中全量嵌入隐形水印，可在复制粘贴和适度编辑后保留，用于标记 AI 生成文本。此举受欧盟《AI 法案》推动，覆盖 API、Claude 等全产品线及云渠道，并计划开发检测工具供用户查验。
正文摘录
- title: Claude Launches "Invisible Watermark" Amid Criticism: New Models Embed It Across the Board, Marking All Text - sourcecompany: QbitAI - bodymarkdown: It will be copied and pasted along with the text to other places, and may still survive after a certain degree of editing. The direct driver behind this move is the EU AI Act, which Anthropic has signed. This mechanism will apply to all models released on or after August 2, 2026. For Claude models currently in production, Anthropic is also researching how to retroactively add this watermarking mechanism. The guidelines were drafted by independent experts, evaluated by the European Commission and the AI Office, and have been signed by around 190 organizations, including Google, Meta, Microsoft, and OpenAI. When Claude generates support…