Claude 的文本水印是怎么工作的
Future Claude models will generate text that contains a watermark. This is a way of determining the likelihood that Claude was involved in writing the text, and we, along with several other major AI providers, are implementing this change to comply with the EU AI Act.
往后的 Claude 模型生成的文本里会带水印。靠它可以判断一段文字有多大可能是 Claude 参与写的。我们和另外几家主要的 AI 厂商一样,做这个改动是为了符合欧盟《人工智能法案》(EU AI Act)。
In this article, we share answers to some of the questions we’ve received about how our chosen watermarking method works, whether it affects Claude’s outputs, and why we’re making this change. To summarize:
这篇文章回答我们收到的一些问题:我们选的水印方法是怎么工作的,会不会影响 Claude 的输出,以及为什么要这么做。先说结论:
- We use a method of watermarking that does not have any practical impact on the quality or content of Claude’s outputs;我们用的水印方法,对 Claude 输出的质量和内容没有任何实际影响;
- The difference between watermarked and un-watermarked text will not be distinguishable to readers;读者分不出哪段文字带水印、哪段不带;
- Nothing is added to the text and there are no hidden characters;文本里什么都没加,也没有隐藏字符;
- Watermarking doesn’t require extra tokens, and will not be more expensive;水印不需要额外的 token,不会更贵;
- Watermarking carries no identifying information and can’t be traced to a specific person, organization, or chat;水印里不带任何身份信息,追溯不到具体的个人、组织或某次对话;
- Watermarking won’t be specific to Claude. As of August 2, the EU requires AI providers serving its market to mark AI-generated content. Other major model developers have signed the same Code of Practice and will be implementing their own watermarks.水印不是 Claude 独有的。从 8 月 2 日起,欧盟要求在它的市场上提供服务的 AI 厂商给 AI 生成的内容打标记。其他几家主要的模型开发商也签了同一份《行为准则》(Code of Practice),也会上线各自的水印。
What is watermarking?水印是什么?
Large language models like Claude work by generating one word at a time. Each time the model decides on the next word, it chooses among a list of potential candidates, ultimately selecting the most sensible or likely based on the preceding text. Take the sentence “The weather today was cold and…”. The next word is very unlikely to be “sugary.” But it is quite likely to be “overcast” or “grey.” Under most circumstances, it doesn’t matter much to the reader which of these latter two words the model ultimately chooses—the meaning of the sentence is largely the same either way. In cases like this, the choice is settled by a random number.
Claude 这样的大语言模型是一次生成一个词的。每次决定下一个词,模型都会从一串候选词里挑,最后根据前文选出最合理、可能性最高的那个。拿「今天天气很冷,而且……」这句来说,下一个词基本不可能是「甜腻」,但很可能是「阴沉」或者「灰蒙蒙」。大多数时候,模型最后选这两个里的哪一个,对读者来说没多大区别,句子的意思差不多。像这种情况,选哪个是由一个随机数决定的。
Watermarking uses low-stakes choices like these—which occur many times over a piece of generated text—to leave a pattern in Claude’s responses. That pattern is undetectable to the reader, but is detectable to anyone who has a key that encodes it. When watermarking is used, choices are still made at random, but the source of the randomness is different. Instead of using an arbitrary random number generator to pick the next word, watermarking uses the key and a few words that come before to settle what word the model should pick. That is, the words that Claude picks are still random, but now, one can check the sequence of words and see if it’s consistent with the choices Claude would make if it was using the key. If it is, one can assign a probability that the text was generated by Claude.
水印就是利用这种无关紧要的选择(一段生成的文本里会出现很多次),在 Claude 的回答里留下一种模式。读者察觉不到这种模式,但手里有对应密钥的人能检测出来。加了水印之后,选择仍然是随机的,只是随机性的来源变了:不再用一个任意的随机数生成器来挑下一个词,而是用密钥加上前面几个词来决定模型该选哪个。也就是说,Claude 挑的词还是随机的,但现在可以拿这串词去核对,看它和 Claude 用这把密钥时会做出的选择对不对得上。对得上的话,就能算出这段文字由 Claude 生成的概率。
Importantly, it isn’t that the model will now always be biased toward overcast or grey. Just as with non-watermarked text, overcast might be selected in one sentence, grey in the next, depending on the words that came before. And it’s not the case that the watermarking method pushes Claude to choose a word it wouldn’t have considered anyway (for instance, it wouldn’t make Claude pick a word like “nubilous”—an obscure1 synonym for overcast or grey that Claude almost certainly wouldn’t use under normal circumstances).
要说明一点:这不等于模型从此总偏向「阴沉」或「灰蒙蒙」。和不带水印的文本一样,这一句可能选了「阴沉」,下一句可能选「灰蒙蒙」,看前面是什么词。水印方法也不会逼 Claude 去选一个它本来就不会考虑的词(比如它不会让 Claude 选「nubilous」这种词。这是「阴沉」「灰蒙蒙」的一个生僻1同义词,正常情况下 Claude 几乎肯定不会用)。
How does watermarking affect Claude’s outputs?水印会怎样影响 Claude 的输出?
Watermarking does not impact the quality of Claude’s output. To a reader, a watermarked response is indistinguishable from an unwatermarked one (in this way, AI watermarks differ substantially from their namesakes on banknotes, other physical objects, and some digital documents, which are visible to the naked eye).
水印不影响 Claude 输出的质量。在读者眼里,带水印的回答和不带水印的没有区别(这一点上,AI 水印和它的同名物很不一样:钞票、其他实物和一些电子文档上的水印都是肉眼看得见的)。
In internal testing, we’ve seen no impact of watermarking on the content, level of creativity, or readability of Claude’s text. In the SynthID-Text paper, which introduced the technique we use, Google DeepMind tested this impact by serving a model that used watermarking to a portion of their Gemini traffic and comparing thumbs-up and thumbs-down ratings. They found no statistically significant differences from the unwatermarked model. And in a controlled study, human raters comparing watermarked and unwatermarked answers side-by-side saw no difference in quality.
我们内部测试下来,水印对 Claude 文本的内容、创造力和可读性都没有影响。我们用的技术出自 SynthID-Text 论文,Google DeepMind 在论文里测过这种影响:他们把一个带水印的模型放到一部分 Gemini 流量上,对比用户点赞和点踩的情况,结果和不带水印的模型相比没有统计显著的差异。另外在一项对照研究里,人工评审把带水印和不带水印的回答并排比较,也没看出质量上有差别。
A useful analogy is to imagine you’re playing a game like Monopoly. On each turn, each player moves a random number of spaces around the board according to the roll of a die. Suppose that, instead of rolling the die to get this randomness, we decided to use a book of the digits of pi.2 We start from a randomly-chosen digit (say, the 1,012,845th after the decimal place, which happens to be a 6), and from that point on each player simply uses the next digit in the sequence as their next “roll."
打个比方,假设你在玩大富翁(Monopoly)这类游戏。每一轮,每个玩家按掷骰子的点数在棋盘上走随机的步数。现在换个玩法,不掷骰子了,改用一本印满圆周率各位数字的书来提供随机性。2从随机挑的某一位开始(比如小数点后第 1,012,845 位,恰好是 6),之后每个玩家都直接拿序列里的下一位数字当自己这一轮「掷」出的点数。
For all intents and purposes, the moves are still random: it makes no difference to the players—or to the outcome of the game—whether the randomness comes from pi or from dice rolls each time. But if we could see the sequence of all the moves after the game (and we knew the value of pi), we could work out whether this was a game that likely used pi to determine its moves. The game that used pi is, in a sense, “watermarked."
不管怎么看,这些步数仍然是随机的:每次的随机性来自圆周率还是骰子,对玩家、对游戏结果都没有任何区别。但要是游戏结束后我们能看到全部走法(而且知道圆周率的值),就能推断出这局游戏是不是很可能用圆周率来决定步数。从某种意义上说,用圆周率玩的那一局就被「打上了水印」。
It’s the same for Claude-generated text. Watermarking doesn’t change the meaning or experience for the person reading it, but if you wanted to check after the fact whether the text was likely generated by Claude, the watermark allows you to do so.
Claude 生成的文本也是这样。水印不会改变读者看到的意思和阅读体验,但如果你事后想查一段文字是不是很可能出自 Claude,水印让你查得出来。
Which specific method of watermarking do you use?你们具体用的是哪种水印方法?
Claude’s text watermark is a version of the SynthID-Text approach published by Google DeepMind in a Nature paper in 2024. It belongs to a family of approaches that go back to a proposal by Scott Aaronson in 2022, all of which share the same design principle that we described above—the watermark only changes the source of the randomness used to pick among words.
Claude 的文本水印是 SynthID-Text 方法的一个版本。这个方法是 Google DeepMind 2024 年在一篇 Nature 论文里发表的,属于一类可以追溯到 Scott Aaronson 2022 年那个提案的方法。这类方法的设计原则都和上面讲的一样:水印只改变挑词时所用随机性的来源。
There are limitations to the effectiveness of watermarking. Using our key, one can only answer the question “What is the likelihood this was partly written by Claude?” It doesn’t confirm whether the text was human-written, and it can’t tell whether the text was written by a different AI (even if that other AI uses watermarking, it would have a different key; it might also use a different watermarking method altogether). Detecting a watermark also doesn’t work well on small samples, where there are fewer word choices and thus less information to go on. As a passage increases in length, confidence about Claude’s involvement increases too.
水印的效果是有局限的。用我们的密钥,只能回答「这段文字有多大可能部分出自 Claude?」这一个问题。它没法确认文字是不是人写的,也分辨不出是不是别的 AI 写的(就算那个 AI 也用水印,密钥也不一样,甚至可能用的是另一套水印方法)。样本很短时,水印检测的效果也不好,因为能选的词少,可依据的信息也就少。段落越长,判断 Claude 参与过的把握就越大。
Watermarking is sparser on factual passages where there are fewer choices that can be made without decreasing the accuracy of the text. For example, take the sentence “Isaac Newton’s most famous work was called Principia …”. It really matters whether the next word is “ Mathematica ” (it’s the only right answer), so the watermark would have nothing to act on. The same is true for proofreading. If you hand Claude a piece of writing and ask it to edit only the grammar and punctuation and nothing else, the watermark can only live in the handful of corrections, which might be too few to register.
讲事实的段落里,水印会更稀疏,因为在不损害准确性的前提下能做的选择更少。比如「艾萨克·牛顿最著名的著作叫 Principia……」这句,下一个词是不是「Mathematica」非常要紧(这是唯一正确的答案),所以水印在这里无处下手。校对也一样。如果你把一段文字交给 Claude,让它只改语法和标点、别的都不动,水印就只能落在那寥寥几处修改上,可能少到检测不出来。
What about cases where Claude has proofread or edited human text?Claude 校对或编辑过的人写的文字呢?
The watermark only applies to words Claude chooses. When Claude proofreads text written by a person, what it gives back has generally only been lightly edited; because nearly all the words are the person’s, there’s very little (if anything) for the watermark to attach to. Depending on the length of the text and how heavily Claude has edited it, those changes might not be enough to make Claude’s involvement detectable. The more Claude writes, the more decisions it has to make, and the more space there is for a watermark.
水印只作用于 Claude 自己选的词。Claude 校对人写的文字时,交回来的一般只是轻度改动过的版本;几乎所有的词都还是作者本人的,水印能附着的地方很少(甚至没有)。看文本有多长、Claude 改动有多大,这些修改可能不足以让人检测出 Claude 参与过。Claude 写得越多,要做的决定就越多,水印的空间也就越大。
What about code?代码呢?
As we noted above, AI watermarking takes advantage of decisions where either choice of a word would be equally good. Where an exact output is required—where there isn’t a choice, and something would be factually wrong or a piece of code would break if a different term was chosen—the watermark isn’t applied.
前面说过,AI 水印利用的是那些选哪个词都一样好的决定。需要精确输出的地方,也就是没得选、换个词事实就错了或者代码就跑不起来的地方,水印不会生效。
For example, once the model has written “2 + 2 =”, there is a very clear best choice for the next token (if the model is completing the sum, there isn’t an answer that’s equally as good as “4”; if it’s talking about George Orwell’s Nineteen Eighty-Four, there isn’t an answer that’s equally as good as “5”). The “nudge” of the watermark wouldn’t be applied here. For the same reason, code—which in very many cases has to be exact—has generally less watermarking than some other forms of text.
比如模型写下「2 + 2 =」之后,下一个 token 有非常明确的最佳选择(如果模型是在算这道加法,没有哪个答案能和「4」一样好;如果它在聊乔治·奥威尔的《一九八四》,也没有哪个答案能和「5」一样好)。水印那一下「轻推」在这里就不会施加。出于同样的原因,代码在很多情况下都必须精确,所以通常比其他一些类型的文本带的水印少。
Having said that, in areas where there is an arbitrary choice between particular words or terms within the code, the watermark can be used, such as comments within code. But by definition, it will have a negligible effect on the actual code produced.
话虽如此,代码里确实存在可以随便选词或术语的地方,比如代码注释,水印还是能用上。但按定义,它对实际生成的代码影响微乎其微。
What does this mean for users?这对用户意味着什么?
Does this slow the model down, or make it more expensive?会让模型变慢或变贵吗?
No. Watermarking has a negligible impact on the speed of models, and because it produces no extra tokens, the model is the same price to serve and use.
不会。水印对模型速度的影响可以忽略不计,而且它不产生额外的 token,所以模型的服务成本和使用价格都不变。
Can a watermark be traced back to me or my organization?水印能追溯到我或我的组织吗?
No. The watermarking applies to Claude and its outputs. It doesn’t identify anything to do with individual users. There’s nothing in the watermark, or its key, that would allow anyone to recover any information about the user, their organization, or their chats with Claude.
不能。水印作用于 Claude 和它的输出,不会标识出任何跟具体用户有关的东西。水印本身和它的密钥里,都没有任何东西能让人还原出用户、用户所在的组织或者他们和 Claude 的对话的信息。
Why are you watermarking Claude’s outputs?你们为什么要给 Claude 的输出加水印?
We’re implementing watermarking to comply with the EU AI Act. Anthropic, along with several other major AI model providers and around 190 total signatories, signed the EU Code of Practice on Transparency of AI-Generated Content in July 2026. This requires AI system providers to use methods of “marking” AI-generated text. We’re applying watermarking globally at launch because we don't yet have a durable way to scope it by region. However, we will continue to evaluate different approaches, and will share updates when we have them.
我们加水印是为了符合欧盟《人工智能法案》。2026 年 7 月,Anthropic 和其他几家主要的 AI 模型厂商,连同总共约 190 个签署方,签署了欧盟的《AI 生成内容透明度行为准则》(Code of Practice on Transparency of AI-Generated Content)。准则要求 AI 系统提供方用某种方法给 AI 生成的文本「打标记」。上线时我们会在全球范围启用水印,因为目前还没有一个可靠的办法按地区划定范围。不过我们会继续评估不同的做法,有进展再跟大家同步。
Other questions其他问题
How do I check if a piece of text was written by Claude?怎么查一段文字是不是 Claude 写的?
We are releasing a detection API in private preview. It is currently available to eligible organizations as required under EU law (such as regulators, law enforcement, media, fact-checkers, independent researchers, educational organizations, and EU civil society groups). It is also available for enterprises who are similarly obligated to verify watermarking for their own compliance with the Act. We plan to expand access to the detection API over time. You can register interest in access here.
我们正在以内测(private preview)的形式推出一个检测 API。按欧盟法律的要求,目前向符合条件的组织开放(比如监管机构、执法部门、媒体、事实核查机构、独立研究者、教育机构和欧盟的民间社会团体)。同样需要核验水印、以满足自身法案合规要求的企业也可以用。我们打算逐步扩大检测 API 的开放范围。想申请的话可以在这里登记。
What about images and other files?图片和其他文件呢?
When Claude produces a file of a supported type (such as a .png, .jpg, or .svg), it will attach a content credential in the form of a small, cryptographically signed note in the file’s metadata, saying that the file was made or processed with Claude. This is an open industry standard called C2PA —the same used by camera manufacturers and in photo-editing software to record where an image came from. Any C2PA-aware tool can read it; we’ll be providing our own where you can drop a file and check.
Claude 生成受支持类型的文件(比如 .png、.jpg 或 .svg)时,会附上一份内容凭证(content credential):文件元数据里一小段带加密签名的说明,写明这个文件是用 Claude 制作或处理的。这是一个叫 C2PA 的开放行业标准,相机厂商和修图软件也用它来记录一张图片的来源。任何支持 C2PA 的工具都能读取它;我们也会提供自己的工具,把文件拖进去就能查。
This metadata label is very different from a watermark. Nothing in the file changes—it is not embedded or hidden. As with text, the credential only says Claude was involved in producing the file; it doesn’t include any identifying information.
这种元数据标签和水印很不一样。文件内容本身一点没变,凭证既没有嵌进去,也没有藏起来。和文本一样,凭证只说明 Claude 参与了文件的制作,不包含任何身份信息。
Can’t someone just edit the text to get around the watermarking?改改文字不就能绕过水印了吗?
To some extent, yes. Light editing probably won’t remove the watermark completely; a complete rewrite where every word is replaced will. In the latter case, of course, it’s arguable whether the text can any longer be described as AI-generated.
某种程度上可以。轻度编辑大概去不干净水印;把每个词都换掉的彻底重写就能去掉。当然,到了后一种情况,这段文字还算不算 AI 生成的,本身就值得商榷了。
What does a watermark actually prove?水印到底能证明什么?
A watermark can only determine that Claude was likely involved with the content at some point. It cannot distinguish “Claude wrote this” from “Claude heavily edited this.”
水印只能判断 Claude 很可能在某个环节参与过这段内容。它分不清「这是 Claude 写的」和「这是 Claude 大改过的」。
Do watermarks apply to translations?翻译也带水印吗?
Yes. A translation produced by Claude carries a watermark, because in this case every word is chosen by Claude.
带。Claude 产出的译文会带水印,因为这种情况下每个词都是 Claude 选的。
What about older Claude models?旧版 Claude 模型呢?
The EU law includes a transition period for Anthropic models launched before August 2, 2026, and we’re working to add watermarking for those models as well. This will be rolled out over the coming months.
对 2026 年 8 月 2 日之前发布的 Anthropic 模型,欧盟法律留了过渡期,我们也在给这些模型加水印,接下来几个月会陆续上线。
How does this differ from AI detection software, like Pangram?这和 Pangram 这类 AI 检测软件有什么不同?
AI detection software uses a different method, because the companies that provide it don’t have our key. Among other things, those services look at aspects of the text like the subtle (and not-so-subtle) “tells” that often appear in AI’s phrasing. For example, AI models appear to be fond of the construction “this isn’t [X], it’s [Y]”, and use the word “quietly” a lot more than you might expect. Picking up on these patterns is fundamentally different from checking for a watermark.
AI 检测软件用的是另一套方法,因为做这些服务的公司没有我们的密钥。它们会看文本的各种特征,其中之一就是 AI 措辞里常冒出来的那些或隐或显的「破绽」。比如 AI 模型好像特别爱用「这不是 [X],而是 [Y]」这种句式,「quietly」(悄悄地)这个词也用得比你想象的多得多。捕捉这类模式,和检查水印是根本不同的两回事。
Does this change who owns a given output, or who is legally responsible for it?这会改变某个输出归谁所有、由谁承担法律责任吗?
No. A watermark only helps test whether Claude might have produced or processed the content. It doesn’t say anything about ownership or authorship, and doesn’t change a user’s rights under our terms. We only apply the watermark when Claude was involved in processing the content or file.
不会。水印只是帮忙检验 Claude 有没有可能产出或处理过某段内容,它不涉及所有权或作者身份,也不改变用户在我们条款下的权利。我们只在 Claude 参与处理了内容或文件时才加水印。
Updated Sep 1, 2026: Provided up to date information on the watermarking detection API.
*2026 年 9 月 1 日更新:*补充了水印检测 API 的最新信息。
1 Or, you might say, nubilous—which is also a synonym for “obscure.”
1 或者说,nubilous 这个词本身就挺 nubilous 的,因为它也有「晦涩」的意思。
2 Pi is technically predictable, but any run of digits from somewhere in the middle of pi is indistinguishable from a run of rolls of a ten-sided die. Also, set aside the fact that the dice in Monopoly go from 1 to 6 whereas a digit of pi can be from 0 to 9; the analogy isn’t perfect.
2 严格来说圆周率是可预测的,但从圆周率中间任取一段数字,和掷十面骰子得到的一串点数是分不出来的。另外,先别管大富翁的骰子是 1 到 6、而圆周率的一位数字是 0 到 9 这回事,这个比方并不完美。