{"id":818108,"date":"2026-08-18T22:22:23","date_gmt":"2026-08-18T22:22:23","guid":{"rendered":"https:\/\/www.newsbeep.com\/us\/818108\/"},"modified":"2026-08-18T22:22:23","modified_gmt":"2026-08-18T22:22:23","slug":"anthropics-text-watermarking-proves-ai-companies-do-not-care-at-all-about-writing","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/us\/818108\/","title":{"rendered":"Anthropic\u2019s Text Watermarking Proves AI Companies Do Not Care at All About Writing"},"content":{"rendered":"

Earlier this month, Anthropic announced that future versions of Claude will generate text that includes watermarks showing it was AI-generated. At the time, Anthropic did not explain how this would work, leaving us to speculate on the podcast: Would it somehow encode this into the text? Include invisible characters? Do something with the metadata? We now know, thanks to a blog post over the weekend, that Anthropic will do this by changing how its AI writes altogether.\u00a0<\/p>\n

\u201cNothing is added to the text and there are no hidden characters,\u201d Anthropic wrote in that company blog post<\/a>. \u201cThe difference between watermarked and un-watermarked text will not be distinguishable to readers.\u201d The way it will work, the post explained, is that Anthropic will subtly alter the word choices in AI-generated text in a way that is only known to Anthropic and its algorithms. Anthropic will know the watermarking algorithm, which will change \u201cthe source of the randomness used to pick among words\u201d and thus can write a tool to detect whether something has been AI-generated.<\/p>\n

This research and approach is interesting in a data science kind of way, but Anthropic\u2019s layperson explanation for how this will work shows how little the company thinks about the craft of writing or the subtle differences between words a human author might want to use to convey their thoughts.\u00a0<\/p>\n

Anthropic asks us to consider the difference between two sentences: \u201cTake the sentence \u2018The weather today was cold and\u2026\u2019. The next word is very unlikely to be \u2018sugary.\u2019 But it is quite likely to be \u2018overcast\u2019 or \u2018grey.\u2019 Under most circumstances, it doesn\u2019t matter much to the reader which of these latter two words the model ultimately chooses\u2014the meaning of the sentence is largely the same either way. In cases like this, the choice is settled by a random number,\u201d Anthropic writes. \u201cWatermarking uses low-stakes choices like these\u2014which occur many times over a piece of generated text\u2014to leave a pattern in Claude\u2019s 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.\u201d<\/p>\n

Anyone who has written anything would, I hope, understand that the difference between the sentences \u201cThe weather today was cold and grey\u201d and \u201cThe weather today was cold and overcast\u201d are sometimes \u201clow stakes,\u201d as Anthropic describes, but not always. \u201cGrey,\u201d and \u201covercast\u201d are different words, and there are any number of reasons why a human author might pick one over the other in a given context. In this example, however, Anthropic\u2019s algorithm sees these words as totally interchangeable and thus its watermarking algorithm has decided that it can \u201cnudge\u201d the word choice one way or the other for the purposes of watermarking.\u00a0<\/p>\n

Anthropic continues: \u201cInstead 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\u2019s 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.\u201d<\/p>\n

Anthropic claims \u201cWatermarking does not impact the quality of Claude\u2019s output. To a reader, a watermarked response is indistinguishable from an unwatermarked one,\u201d and that \u201cin internal testing, we\u2019ve seen no impact of watermarking on the content, level of creativity, or readability of Claude\u2019s text.\u201d<\/p>\n

People are quite mad about Anthropic\u2019s watermarking system, and understandably so. Synonyms are sometimes interchangeable, but not always, as is pointed out in this excellent essay by John Gruber of Daring Fireball<\/a>, and by journalism academic Jeff Jarvis, in which he claims Anthropic \u201cdevalues writing<\/a>.\u201d In making this choice, \u201cAnthropic declares words fungible, language random, choice meaningless,\u201d Jarvis writes.\u00a0<\/p>\n

When I sat down to write this post, I was mad because it seems like Anthropic is\u00a0 putting its thumb on the scale, messing with the outputs of its machine and saying that the resulting text is qualitatively just the same as the other AI text it was probably going to output. But as I began writing this, I realized that my problem is not necessarily with text watermarking but with AI-generated text altogether. It does not matter to me, necessarily, whether the output of Claude\u2019s garbage AI text is one way or is a slightly different way. But it does matter to me that AI data scientists at huge tech companies think that word choice doesn\u2019t matter, or that it is possible to statistically use synonyms wherever without fucking with the meaning of a sentence.\u00a0\u00a0<\/p>\n

Throughout the blog post, Anthropic describes the act of writing as being akin to a probabilistic game of chance. In Anthropic\u2019s own words, its writing is sometimes the result of an \u201carbitrary random number generator,\u201d and \u201crandom\u201d whenever its systems encounter a situation where its tool believes, based on pattern recognition, that the choice between several possible next words isn\u2019t all that important. That may be true for LLM garbage, but is not true for the human experience of writing, which is why human writing almost always feels different than AI writing.<\/p>\n

This watermarking approach, and Anthropic\u2019s blog post about it, highlights something that should already be clear about a company that famously scanned and destroyed huge numbers of printed books and has trained its LLMs on stolen content: Anthropic does not care about the craft or effort of writing, and sees words as fungible and unimportant. Anthropic says it is making this change as part of the European Union\u2019s new AI regulations, which are well-intentioned but problematic. While it can definitely be useful to have additional ways of detecting AI-generated content, the carelessness with which Anthropic has announced this decision highlights the broader problem with using LLMs to write: They are, as Anthropic notes, probabilistic tools that do not \u201cwrite\u201d in the way that humans do, rather, they mimic their training data which is, by definition, things that have already happened and been ingested.\u00a0<\/p>\n

Contrast this with how Anthropic sees code, something where it says an \u201cexact output is required.\u201d In writing, meanwhile, Anthropic suggests different words are often \u201cequally good.\u201d Over and over again, Anthropic and the researchers who work on this type of watermarking claim that text can be \u201cnudged\u201d in this way without being noticeable to humans or without impacting \u201cquality.\u201d\u00a0<\/p>\n

But it is worth noting that the people judging the \u201cquality\u201d of the AI-generated outputs are either data scientists or people asking AI tools to do their writing for them, not, say, people who care about reading or writing. The scientific paper that Anthropic cites was done by Google researchers<\/a> on a Google watermarking tool called \u201cSynthID,\u201d which Anthropic\u2019s watermarking is based on.\u00a0<\/p>\n

In the SynthID study, quality was assessed by randomly putting watermarking on some Gemini outputs, then asking Gemini users to either thumbs-up or thumbs-down the response: \u201cA random fraction of queries were routed to a watermarked model and an equivalent number to the unwatermarked counterpart. The Gemini user interface allows users to provide feedback on model responses via a thumbs-up (good response) and a thumbs-down (bad response). We analysed approximately 20\u2009million watermarked and unwatermarked responses and computed the thumbs-up and thumbs-down rates (both as a fraction of the total number of thumbs-up and thumbs-down feedback received). We found that the thumbs-up rate for the two models differed by 0.01%.\u201d<\/p>\n

I hope it is clear to anyone who has clicked on this article that asking someone who asked a chatbot something to thumbs up or thumbs down a response is not a very good way of assessing the \u201cquality\u201d of \u201cwriting.\u201d The other human assessment that Google did was to ask people to assess side-by-side watermarked and unwatermarked text for quality. Here are examples given in an appendix of the study; apparently people did not really have a preference one way or the other:<\/p>\n

\"\"<\/p>\n

One could argue that these passages are two different ways of explaining something, yes. But they are definitively not the \u201csame,\u201d and it is unclear to any reader why one version is one way and the other version is another way. Why did the LLM write \u201crespiratory failure\u201d in one example and \u201ccessation of breathing\u201d in the other? The answer for both is an \u201carbitrary random number generator\u201d and proprietary black box algorithmic weighting systems controlled by the AI company. In the watermarked version there\u2019s been an additional \u201cnudging\u201d or messing with the machine that\u2019s already just a pattern matcher.\u00a0<\/p>\n

The point is, there is no conscious thought or decision-making process happening here, so perhaps watermarked AI text is not all that much more offensive than regular AI text. But to see it laid out in such stark terms by the companies building these machines shows how little they actually care about writing. If you asked me, on the other hand, why I used one word instead of another, I might not be able to tell you exactly why, but I could probably explain to you what I was going for, the style of writing I do, my intended audience, my mood that day, whether my heart was racing or not, where I was, what I was doing, what I did earlier that morning and what I did later that day. Maybe it was a word my third grade teacher used all the time or which I read in an article last week or is an inside joke with my friends or which I have recently become obsessed with or tend to overuse. Why I wrote what I wrote or why I did anything at all is the result of my some mix of human experiences dating back to when I first acquired language as a baby and continuing on to this very moment that I may or may not be able to explain, but which result in a certain style of writing that is mine.<\/p>\n

This is the case even when I\u2019m working fast or carelessly dashing off text messages, when the thoughts just kind of flow from my brain to my fingers to my keyboard where I don\u2019t know if what I\u2019m saying is making sense at all but is probably legible because it\u2019s coming from a human brain and not a random number generator.\u00a0<\/p>\n

This is why short passages of AI-generated text feel soulless and generic<\/a>, as we have written about repeatedly. And there are many AI tools that use AI to make AI writing seem less generic (yo dawg, we heard you like AI so we put AI in your AI) by using synonyms that are supposed to make a passage sound more human \u2014 or less plagiarized \u2014 by picking words that are less commonly used. The text outputted by these tools, which are called \u201cspinners\u201d or \u201chumanizers\u201d are often just as uncanny and weird as AI writing itself. Or, when applied to things where, to use Anthropic\u2019s own language, \u201can exact output is required\u201d such as quotes in a news article, the output is often factually inaccurate, libelous, or just plain garbage.\u00a0\u00a0<\/p>\n

About the author<\/p>\n

Jason is a cofounder of 404 Media. He was previously the editor-in-chief of Motherboard. He loves the Freedom of Information Act and surfing.<\/p>\n

\"Jason <\/p>\n","protected":false},"excerpt":{"rendered":"Earlier this month, Anthropic announced that future versions of Claude will generate text that includes watermarks showing it…\n","protected":false},"author":2,"featured_media":818109,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[45],"tags":[182,181,507,74],"class_list":["post-818108","post","type-post","status-publish","format-standard","has-post-thumbnail","category-artificial-intelligence","tag-ai","tag-artificial-intelligence","tag-artificialintelligence","tag-technology"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/818108","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/comments?post=818108"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/posts\/818108\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media\/818109"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/media?parent=818108"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/categories?post=818108"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/us\/wp-json\/wp\/v2\/tags?post=818108"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}