[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-anthropic-claude-text-watermark-eu-ai-act":3,"news-related-ca53004e-9180-4b9d-b9db-337f2d20994b":41},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":27,"news_slug":34,"published_at":35,"created_at":36,"modified_at":37,"is_published":38,"publish_type":39,"image_url":14,"view_count":40},"ca53004e-9180-4b9d-b9db-337f2d20994b","Anthropic 给 Claude 文本加水印:欧盟 AI Act 第 50 条第一次有了「出厂级」答案","Anthropic 从 2026 年 8 月 2 日起为新 Claude 模型引入不可见文本水印和 C2PA 文件来源签名,以满足欧盟 AI 法案第 50 条透明度义务,并已签署配套的欧盟 AI 实践准则。水印在模型层注入,覆盖 API、Claude、Claude Code、Claude Cowork、Claude Tag 及 AWS、Google Cloud、Microsoft Foundry 等云端渠道,适用全球用户而非仅限欧盟;旧模型将在过渡期内陆续补齐。Anthropic 同时承诺公开检测工具的技术细节,并明确指出有标不等于 AI 原创、无标不等于人类撰写。","# Anthropic 给 Claude 文本加水印:欧盟 AI Act 第 50 条第一次有了\"出厂级\"答案\n\n2026 年 8 月 2 日起,Anthropic 为新发布的 Claude 模型引入\"机器可读、人类不可见\"的水印机制,覆盖 API、Claude、Claude Code、Claude Cowork、Claude Tag 等所有支持产品,云端用户通过 AWS、Google Cloud、Microsoft Foundry 接入的 Claude 输出同样会被标记。Anthropic 已签署欧盟《AI 实践准则》(Code of Practice),这次更新的直接合规对象是《欧盟 AI 法案》第 50 条——AI 生成内容的透明度义务。[^1]\n\n但要点不只是\"又一家厂商加了水印\"。这次有两件事不一样:**水印在模型层注入,而不是在产品层贴标**,而且**适用范围是全球用户,不止欧盟**。\n\n## 两套机制,文本与文件分开处理\n\nAnthropic 没用单一方案,而是把文本和文件分到两条完全不同的技术轨道上:\n\n- **文本:不可见水印**。在模型生成阶段,把一个人类阅读时感知不到的信号直接编进 token 序列里。用户正常阅读时不会察觉到任何变化,但复制粘贴后,水印仍然跟着文本走;Anthropic 还预计**部分水印在用户编辑后仍可被检出**——这一点对\"改两句就绕过去\"的常见质疑是关键回应。\n- **文件(含常见图像格式):C2PA 签名来源信息**。文件走 Coalition for Content Provenance and Authenticity 标准,把\"经 Claude 处理\"这件事写到可签名的元数据里,第三方可以校验签名是否被篡改。\n\n这种\"分而治之\"的选择背后有技术上的硬道理。文本水印靠的是 token 序列的统计偏差,跟着内容走、不随格式变化消失,所以对抗\"复制粘贴改写\"比较强。文件元数据则天然怕格式转换、截图、另存——这些操作能把元数据直接剥掉。把两条机制分开,等于让每种载体用最适合自己生存环境的取证手段。\n\n## 为什么是\"模型层\"?\n\nAnthropic 把水印施加点放在**模型层**(in the model itself),而不是某个上层产品或 API 网关。这是这次方案最值得拆解的地方。\n\n如果水印贴在前端,那换个集成方、换个部署形态,标记就丢了——黑产只要调用同一个底层模型换一层包装就能洗掉。把水印做到模型生成的最后一公里,意味着**任何调用这个模型的下游产品**——无论是官方 Claude 客户端、第三方 IDE 插件、还是 AWS 上跑的私有部署——拿到的输出都自带标记。Anthropic 列出的覆盖面包括 API、Claude、Claude Code、Claude Cowork、Claude Tag,以及通过 AWS、Google Cloud、Microsoft Foundry 等云平台接入的客户,这种\"凡是调用就带标\"的覆盖度,只有在模型层做才能成立。\n\nAnthropic 还承诺会**公开发布检测水印的技术细节**,让用户和第三方可以自行验证内容是否带标。这步棋的潜台词是:不要把\"有标 \u002F 无标\"当成 Anthropic 一家说了算的黑箱,把检测工具开放出来,才有第三方生态——也才有真正可被审计的透明度。\n\n## 合规驱动力:欧盟 AI Act 第 50 条\n\n直接的合规驱动力是**《欧盟 AI 法案》第 50 条**关于 AI 生成内容透明度的要求。Anthropic 已经签署了配套的**欧盟《AI 实践准则》**,自愿承诺按第 50 条的要求给生成内容打标。\n\n值得注意的是,水印机制**面向全球用户开放**,而不只是欧盟用户。这规避了\"水印版本只在欧洲能用,其他地区继续裸奔\"的合规套利——只要你的输出经过 Claude,就有标记,没有地域例外。\n\n旧版 Claude 模型**暂时不会被立刻打标**。Anthropic 表示会在欧盟 AI 法案的过渡期内逐步给老模型补上标记能力。这意味着短期内市场上仍会有一段\"新模型带标、旧模型裸奔\"的混合期,对做内容溯源的团队来说是一个需要注意的过渡窗口。\n\n## 不要把水印当成\"作者证明\"\n\nAnthropic 自己也强调了两条**反向警示**——这部分是少有厂商愿意写进官方说明的:\n\n1. **有水印 ≠ Claude 是原作者**。Claude 可以翻译、摘要、改写人类写的稿子,改完之后照样带标。所以\"看到水印 = 这段是 AI 原创\"是错的,正确读法是\"看到水印 = 这段经过了 Claude 的处理\"。\n2. **无水印 ≠ 人类写的**。重度编辑会削弱文本水印;过短的文本可能也缺乏足够的统计信号去稳定识别。C2PA 文件元数据更脆弱,任何格式转换或截图都能让它丢失。\n\n把这两个反向警示摆出来,Anthropic 实际上是在拒绝\"水印即证据\"的简化叙事——他们宁可让公众预期管理做得保守一点,也不愿意日后被反噬。这对一个打算长期做 AI 内容治理的厂商来说,是更稳的姿态。\n\n## 所以呢:监管驱动下,\"出厂级身份\"正在成为新标配\n\n把这次更新放到更大的图景里看,欧盟 AI 法案的透明度条款落地节奏正在从\"原则\"切到\"执行\":大模型厂商开始把**生成内容身份**当作模型本身的一项能力来交付,而不是交给下游应用去补。这对生态有三层影响:\n\n- **对内容平台**:以后接入 Claude 输出会自带机器可读的\"来源指纹\",平台侧的\"AI 内容标注\"工作可以从主动检测退化成被动校验。\n- **对开发者**:在 AWS、GCP、Azure 上调用 Claude 的应用,输出会自动带着可被第三方验证的标记——这降低了做合规审计的工程成本,但也意味着任何想绕开标注的灰色用法会变得更难。\n- **对其他厂商**:如果欧盟 AI Act 第 50 条的执法强度上来,Google、OpenAI、xAI 等迟早要跟进类似方案;现在 Anthropic 把检测工具的技术文档也开放出来,某种程度上是在**把行业标准拉到自己写的位置上**。\n\n水印这件事,过去几年一直停留在研究论文和 demo 阶段。这次 Anthropic 把它推到\"全球所有 Claude 用户、模型层注入、云端全覆盖\"的量产形态,意味着 AI 生成内容的\"出厂身份证\"时代,从监管文本走进了实际流量里。\n\n---\n\n[^1]: Anthropic puts hidden watermarks on Claude text under new EU rules, Interesting Engineering, 2026-08-11. https:\u002F\u002Finterestingengineering.com\u002Fai-robotics\u002Fanthropic-claude-text-invisible-watermarks","https:\u002F\u002Finterestingengineering.com\u002Fai-robotics\u002Fanthropic-claude-text-invisible-watermarks","1fa87d30-d9f3-4752-b3be-0373933b3aaf",[11,15,18,21,24],{"id":12,"name":13,"slug":13,"description":14,"color":14},"c33b1bbc-d6ce-4f61-9d5d-1a0704a6a09b","ai-policy",null,{"id":16,"name":17,"slug":17,"description":14,"color":14},"1fcfaaf2-67de-43d3-9e35-5784852fec60","ai-safety",{"id":19,"name":20,"slug":20,"description":14,"color":14},"23544f6a-eea1-4f05-aa8d-749ca862d5d2","anthropic",{"id":22,"name":23,"slug":23,"description":14,"color":14},"dca4d0ab-7994-43a7-839e-7756fc77344a","claude",{"id":25,"name":26,"slug":26,"description":14,"color":14},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[28],{"id":29,"lang":30,"title":31,"summary":32,"content":33},"9e2b09fc-68b2-4a53-ab78-ad7171ed2824","en","Anthropic watermarks Claude text: a first answer to the EU AI Act","Starting August 2, 2026, Anthropic is rolling out imperceptible text watermarks and C2PA file provenance signatures on new Claude models to comply with EU AI Act Article 50, and has signed the EU Code of Practice covering AI-generated content. The watermark is injected at the model layer and covers the API, Claude, Claude Code, Claude Cowork, Claude Tag, plus cloud customers on AWS, Google Cloud, and Microsoft Foundry, with worldwide availability rather than an EU-only carve-out; legacy models will be brought under the same scheme during the transition period. Anthropic also commits to publishing technical detection documentation, and explicitly cautions that a watermark does not prove AI authorship, nor does its absence prove human authorship.","# Anthropic adds invisible watermarks to Claude text: EU AI Act Article 50 finally gets an \"OEM-grade\" answer\n\nStarting August 2, 2026, Anthropic is rolling out machine-readable, human-imperceptible watermarks on text generated by new Claude models. The mechanism spans the API, Claude, Claude Code, Claude Cowork, and Claude Tag, and applies to cloud customers accessing supported Claude models through AWS, Google Cloud, and Microsoft Foundry. Anthropic has signed the European Union's Code of Practice covering AI-generated content, and the immediate compliance target is Article 50 of the EU AI Act — the transparency obligation for AI-generated content.[^1]\n\nBut two things make this different from \"yet another vendor adds a watermark.\" **The watermark is injected at the model level, not at the product layer**, and **the rollout is global, not EU-only**.\n\n## Two mechanisms: text and files handled separately\n\nAnthropic is not using a single scheme. Text and files ride two technically distinct tracks:\n\n- **Text: an invisible watermark**. At the model generation stage, a signal imperceptible during normal reading is embedded directly into the token sequence. Users do not notice it when reading, and the mark stays attached when they copy and paste responses. Anthropic also expects **some marks to survive editing** — a meaningful response to the standard \"just rewrite two sentences and you're clean\" objection.\n- **Files (including common image formats): C2PA signed provenance**. Files ride the Coalition for Content Provenance and Authenticity standard. Generated files can carry signed provenance metadata that third parties can verify has not been tampered with.\n\nThe split is not arbitrary. Text watermarks rely on statistical biases in token sequences, traveling with the content and surviving format-preserving operations, so they are relatively robust against \"copy, paste, lightly rewrite.\" File metadata, by contrast, is inherently fragile: format conversions, screenshots, and re-saves can strip it. Putting each medium on the technique best suited to its own survival environment is the right engineering trade.\n\n## Why \"model level\" matters\n\nAnthropic applies the watermark **in the model itself**, not at some upper-layer product or API gateway. This is the most important architectural choice in the rollout.\n\nA watermark applied at the front end can be laundered: call the same underlying model through a different wrapper, and the mark disappears. Doing it at the last mile of generation means **any downstream product that calls the model** — the official Claude clients, third-party IDE plugins, even private deployments running on AWS — receives marked output. The coverage Anthropic lists (API, Claude, Claude Code, Claude Cowork, Claude Tag, plus AWS \u002F Google Cloud \u002F Microsoft Foundry customers) is only achievable because the marking happens inside the model.\n\nAnthropic also commits to **publishing technical details for detecting its watermarks**, so users and third parties can verify content on their own. The implicit message: do not treat \"marked \u002F unmarked\" as a one-vendor black box. Open detection tooling is what makes the transparency claim actually auditable.\n\n## Compliance driver: EU AI Act Article 50\n\nThe direct compliance lever is **Article 50 of the EU AI Act** on transparency for AI-generated content. Anthropic has signed the companion **EU Code of Practice**, a voluntary commitment to mark generated content under Article 50.\n\nNotably, the marking system is **available worldwide**, not just to EU users. That forecloses a familiar arbitrage pattern — \"the watermarked build is only available in Europe, everywhere else stays unmarked.\" Anyone whose output passes through Claude gets the mark, with no regional exception.\n\nOlder Claude models are **not immediately marked**. Anthropic says it is working through the EU AI Act's transition period to bring legacy models under the same scheme. In the short term, this creates a mixed fleet — new models marked, legacy models bare — which content provenance teams should plan for.\n\n## Do not treat the watermark as proof of authorship\n\nAnthropic itself flags two **reverse caveats** — rare for a vendor to put in writing:\n\n1. **A watermark does not mean Claude is the author.** Claude can translate, summarize, or edit human-written material and still produce marked output. The correct reading is \"this text was processed by Claude,\" not \"this text is AI-original.\"\n2. **No watermark does not mean a human wrote it.** Heavy editing can weaken a text mark. Short passages may also lack enough signal for reliable detection. C2PA file metadata is even more fragile — any format conversion or screenshot can drop it.\n\nBy spelling these out, Anthropic is refusing the simplified \"watermark equals evidence\" narrative. They would rather set conservative public expectations than deal with the backlash later. For a vendor planning to live in AI content governance for the long haul, that is the steadier posture.\n\n## So what: under regulatory pressure, \"OEM-grade identity\" is becoming the new default\n\nZooming out, the EU AI Act's transparency provisions are moving from principle to execution: frontier model vendors are starting to ship **content identity as a native model capability**, rather than something downstream applications bolt on. That has three layers of ecosystem impact:\n\n- **For content platforms**: Claude output will arrive with a machine-readable \"provenance fingerprint.\" Platform-side \"label AI content\" work can downgrade from active detection to passive verification.\n- **For developers**: Applications calling Claude via AWS, GCP, or Azure get marks that third parties can verify, lowering the engineering cost of compliance audits — but also making gray-area \"wash the mark\" workflows harder.\n- **For other vendors**: If EU AI Act Article 50 enforcement bites, Google, OpenAI, xAI, and others will eventually need similar schemes. Anthropic is also opening the detection-tool technical docs, which in effect pulls the industry standard toward ground they have already staked out.\n\nText watermarking has lived mostly in papers and demos for years. Anthropic is now pushing it into production at the scale of \"every Claude user worldwide, model-level injection, full cloud coverage.\" The era of \"factory-installed identity\" for AI-generated content has moved from regulatory text into actual traffic.\n\n---\n\n[^1]: Anthropic puts hidden watermarks on Claude text under new EU rules, Interesting Engineering, 2026-08-11. https:\u002F\u002Finterestingengineering.com\u002Fai-robotics\u002Fanthropic-claude-text-invisible-watermarks","anthropic-claude-text-watermark-eu-ai-act","2026-08-11T21:48:00Z","2026-08-11T20:08:00.319517Z","2026-08-11T20:08:00.319528Z",true,"agent",415,{"items":42},[43,48,53,58,63,68],{"id":44,"title":45,"news_slug":46,"published_at":47},"97c97b9c-e6e4-4982-aa57-0c0da814fb19","Anthropic 的欧盟答卷四小时即被撕开：Claude 文本水印为什么怕改写","claude-synthid-70-percent-threshold-bypass","2026-08-21T08:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"9f566c9a-4c39-427c-af5e-c3a6b162ec25","Anthropic 把不可见水印写进 Claude 文本：复制粘贴都带走的 AI 身份证","anthropic-claude-invisible-watermark-eu-ai-act","2026-08-12T02:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"470b8663-3916-4bc5-ac3c-c592487c2873","Claude水印官宣4小时被破:开源去除工具走红,水印军备竞赛开场","claude-watermark-removal-tool","2026-08-20T19:30:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"6b07def3-2b1c-4fe9-9922-0dd0038c149c","Anthropic 风险报告更新:Threat Model 1 升至「低」,Threat Model 2 维持「低」但信心下降","anthropic-risk-report-august-2026-update","2026-08-19T03:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"a7f4cfad-874e-42b0-a84b-bd0ec57e8fdc","Anthropic 给 Claude 文本上不可见水印,接 SynthID-Text 走全球合规","anthropic-claude-invisible-text-watermark","2026-08-18T03:30:00+00:00",{"id":69,"title":70,"news_slug":71,"published_at":72},"a124851a-081e-44b6-9f20-f775404279e1","Claude 全球文本水印:Anthropic 把欧盟 AI Act 第 50 条做成\"全球默认\"","claude-text-watermark-eu-ai-act-global","2026-08-14T03:00:00+00:00"]