Claude's AI text watermark sparks backlash, cancellations and a 'perversion of writing' row
Anthropic's EU-mandated watermarking of Claude's text output has triggered subscription cancellations, a stripping tool with 10,000 GitHub stars, and a viral essay calling it a corruption of writing.
What to know
- Anthropic's watermark biases token/word choices during generation to create a statistically detectable but individually invisible pattern, not hidden characters or metadata.
- The rollout, driven by EU AI Act transparency rules, has prompted reported subscription cancellations from users worried about authorship disputes in coding, academic, and client work.
- Critics split between those saying the technique inherently degrades writing quality/creativity and those saying it is technically weak and already being stripped by open-source tools.
- A GitHub tool to remove such watermarks (targeting Claude, SynthID-Text, OpenAI, C2PA and EXIF) passed 10,000 stars within days of Anthropic's technical disclosure.
The dispute Whether the backlash reflects a legitimate concern that watermarking corrupts writing quality and authorship, or is mostly outrage from people upset about being caught using AI. · positions read across 394 posts and comments
Biasing token choice for detectability inherently degrades the writing's quality and creativity, whatever Anthropic claims.
-
“The idea that anything other than *my* needs should factor into the generation of text *for me* is patently offensive.”
John Gruber · Daring Fireball / Mastodon
The real objection is that a hidden marker will expose AI use in professional, client, or academic work, not that it's secretly 'tracking' anyone.
-
“the backlash to Claude’s new text watermark is not about secret tracking but about authorship, quality and who bears the cost of an imperfect detection”
@kimmonismus · X
Watermarks have weak security guarantees and will always be trivial or near-trivial to strip, making the whole scheme largely theater.
-
“Watermarking and fingerprinting have inherently weak security guarantees -- they rely a lot on security through obscurity, weak assumed adversaries to deliver.”
sbszllr · Hacker News
The watermark is reasonable — it can catch plagiarism/cheating and may help prevent models from training on their own AI-generated output.
-
“if it catches "You" who simply copied and pasted an AI response, then that watermark is doing exactly what it's supposed to do”
Ferret51 · Reddit ↗
“It’s unacceptable for a tool to sacrifice an iota of clarity, coherence, meaning, quality, etc. for the purpose of embedding hidden clues within the text to suggest its provenance.”
John Gruber, Daring Fireball writer · Daring Fireball · Aug 16
Anthropic AI lab, maker of ClaudeJohn Gruber Daring Fireball writerArturo Villarroya Claude Max subscriber, public policy workerVladislav Rajtmajer Freelance developer, Czech RepublicJames Padolsey Author of interactive watermarking explainer
How it unfolded 8 developments, newest first · click a bar or a number to jump articlespostscomments
-
8
Guardian and Verge probe whether the watermark hurts quality
The Guardian and The Verge published explainers on the mechanics of the watermark and questioned whether the technique makes AI writing worse, while some commentators pushed back that the outrage over being 'caught' was overblown.
“every needs to go outside, breath, and touch some grass for a few minutes…”
— @ianriccaboni.com -
A
can't quite put my finger on why but this has gotta be the funniest, most absurd, most humiliating sentence i've read in a long long time (from https:// daringfireball.net/2026/08/ant hropics_watermark_text_adulteration_in_claude_is_a_perversion_of_writing )
2 more of the top 3 · 73 posts in this stretch
-
I agree with the substance of this article, and disagree with the author's reaction to it.The part I agree with:It is true that watermarking can be done by "just" swapping one PRNG for another, and it is even true that with today's LLMs, it is possible that this will not degrade the output. But it has a cost, and as things improve, that cost will…
-
J
"a paper in a leading chemistry journal had to be retracted because the authors appeared to have used an AI tool to draft part of the publication. The tool used the phrase “mass killing of an ethnic group” as an alternative for “final solution”, though the paragraph in question appeared to be about a zinc nanogel." # AI https://www…
-
-
7
Gruber calls the watermark 'a perversion of writing'
John Gruber published a widely discussed Daring Fireball essay arguing Anthropic's claim that watermarking doesn't affect quality was false, since biasing word choice for detectability inherently corrupts text generated for the user; it drew a 714-comment Hacker News thread and follow-up Business Insider coverage.
“The idea that anything other than *my* needs should factor into the generation of text *for me* is patently offensive.”
— John Gruber -
first by The Guardian, 61d ago · also Guardian Media, Guardian Tech
-
first by Business Insider, 61d ago
-
K
This is not about copyright, safety or anything human concerns. This is a way to prevent the models from ingesting their own output, and limit scraping to only original human (high value) content. Remember these people buy up old (human authored) books with the container load only to rip them apart to give their models a slight edge. https://…
2 more of the top 3 · 41 posts in this stretch
-
This (Anthropic's "watermark" stance, I mean) is so fundamentally ludicrous that I have assumed it is a (wholly insincere, but arguably pragmatic, at least from their perspective) attempt to deal with the EU and their latest misguided, ham-fisted attempt to solve a real-world problem by drenching the entire world with more regulatory slop[1].The…
-
Future Claude models are getting an invisible text watermark. But not the way you'd think. No hidden characters. No extra tokens. No visible label. No user-identifying information. It's basically a statistical pattern hidden in the model's word choices that Anthropic can check
-
-
6
GitHub watermark-stripping tool passes 10,000 stars
Days after Anthropic detailed the mechanism, an MIT-licensed repository designed to strip Claude, SynthID-Text and OpenAI watermarks (plus C2PA/EXIF metadata) rapidly gained popularity.
“WATERMARK STRIPPER HITS 10,000 STARS DAYS AFTER ANTHROPIC SHIPS PROVENANCE…”
— @AiBreakfast -
C
everybody clearly wants Al which is why if anybody is able to tell a business used it their customers will leave
2 more of the top 3 · 13 posts in this stretch
-
WATERMARK STRIPPER HITS 10,000 STARS DAYS AFTER ANTHROPIC SHIPS PROVENANCE Anthropic detailed how Claude's text watermark works last week. By this morning a MIT-licensed repo that strips it has 10,000 stars, targeting Claude, SynthID-Text and OpenAI marks, plus C2PA and EXIF in
-
B
=> The new Claude's watermarking - from the Anthropic blog: "It doesn’t confirm whether the text was human-written, and it can’t tell whether the text was written by a different AI." # ai # safety # writing # art # images # business # technology # computers https://www. anthropic.com/news/claude-text -watermark
-
-
5
Business Insider reports Claude users canceling over the watermark
Business Insider interviewed four users who said they canceled Claude Max subscriptions specifically citing the watermark, worried it could misattribute authorship on lightly-edited or human-written work in professional and academic settings.
“I do use it for everything, mostly coding apps and tools…”
— Arturo Villarroya -
first by Forbes, 67d ago
-
first by City A.M., 67d ago
-
wonder why! https://www. businessinsider.com/claude-use rs-cancel-subscriptions-citing-anthropic-new-ai-watermark-2026-8
2 more of the top 3 · 28 posts in this stretch
-
E
Also, Anthropic's just the first one out of the gate. The other platforms will be doing this, too, in short order, to comply with the EU AI Act Transparency Code (ref:
-
Anthropic has explained its text watermarking approach.
-
-
4
Anthropic publishes technical explainer for the watermark method
Anthropic released "How Claude's Text Watermarking Works," describing a steganographic technique that biases token selection at inference time to leave a statistically detectable but individually imperceptible pattern.
-
first by The Independent, 61d ago
-
first by TechCrunch, 63d ago
-
J
Watermarking in LLM generated text. Generate a key, and use it to slightly steer word choices, so that a correlation with the key can be observed when compared against the word choices that would be made with the key. Don't forget that LLMs are autocomplete. https://www. anthropic.com/news/claude-text -watermark
2 more of the top 3 · 44 posts in this stretch
-
Interesting. Here's the section of the EU Act that mandates this:> Providers of AI systems, including general-purpose AI systems, generating synthetic audio, image, video or text content, shall ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated. Providers shall…
-
C
Anthropic has revealed that it will soon watermark content that is processed (not just generated!) by any of its models. In a support article, Anthropic explained that it was rolling out machine-readable watermarks to comply with the European Union’s AI Act, which requires all AI system providers to watermark AI-generated or manipulated audio…
-
-
3
Technical critics argue watermarks are trivial to defeat
Software engineer Sean Goedecke published an analysis arguing text AI watermarks will always be easy to remove, and James Padolsey published an interactive essay explaining the statistical word-choice mechanism, both drawing large HN discussions.
-
0
@theverge.com Anthropic says it will put an invisible watermark on #Claude-generated text and images in an effort to comply with European rules for AI transparency.
2 more of the top 3 · 84 posts in this stretch
-
C
🚨 Acaban de crear una skill que elimina las marcas de agua de Claude. Se llama watermarks-remover y está diseñada para limpiar las diferentes señales invisibles que pueden quedar incorporadas en textos y archivos generados con IA. Esto es lo que hace: → Elimina caracteres Unicode invisibles y espacios especiales → Limpia metadatos C2PA, EXIF y…
-
One aspect that people seemingly aren't talking about is the impact this has in the model's creativity. Because the model will nudge each word towards group A vs group B, you're losing on creativity, especially more so if the nudge isn't a gentle 55% but something like 70% or 80%. So essentially they're forcing the model to be less creative for…
-
-
2
Users say the watermark will expose their AI use at work and school
TechCrunch and social media users reacted angrily, worried that a persistent, undetectable-to-them marker could flag work or schoolwork as AI-generated even after editing.
“the backlash to Claude’s new text watermark is not about secret tracking but about authorship, quality and who bears the cost of an imperfect detection…”
— @kimmonismus -
T
Maybe they're extremely clever with half-normalized unicode, but my hunch is: ``` # Ruby text = "[whatever claude gives you]" watermark_removed = text.gsub(/[[:space:]]+/u, ' ').gsub(/[\u200B-\u200D\uFEFF]/,'') ```
2 more of the top 3 · 58 posts in this stretch
-
Will this invisible plaintext # AI “watermarking”work? https:// support.claude.com/en/articles /16266773-how-claude-marks-ai-generated-content # Claude
-
Huh... I wonder if some big version of a bloom filter would work as well. Hash all output text, probably in chunks of a couple tens of tokens each (that would need tweaking to find the most useful hash input length I guess), and smash 'em into a bloom filter. Every month or something, Anthropic releases a giant file containing whatever huge length…
-
-
1
Anthropic announces Claude will watermark AI-generated text
Anthropic published a support document, "How Claude Marks AI-Generated Content," saying supported Claude models would embed an imperceptible watermark in generated text and digitally sign AI-made files/images, to comply with the EU AI Act. The document did not explain the mechanism.
“it weaves an imperceptible watermark directly into the text itself. You won’t see it, and it doesn’t change the meaning, quality, or readability of Claude’s response.”
— Anthropic -
16 outlets first by afaqs!, 67d ago · also BMI, Search Engine Land, SiliconANGLE, PYMNTS, CyberInsider, WinCentral +9 · read ↗
-
10 outlets first by PCWorld, 67d ago · also Quartz, Breitbart, Fortune, CNET, Fast Company, Dawn +3 · read ↗
-
JUST IN: Claude will now invisibly watermark AI-generated text so it can be detected after being copied & pasted.
2 more of the top 3 · 53 posts in this stretch
-
C
To comply with the EU AI Act, Anthropic will now watermark text and photos created by Claude and digitally sign files it creates to indicate they were generated by AI. So going forward text generated by Claude will be detectable as generated by AI programmatically.
-
Scott Aaronson talks about his project at OpenAI to do this^You can carefully select which pseudorandom number generator (prng) you use to be able to id text of a certain length. I expect there is some performance characteristic you have to manage since you're doing this on every inference, but once you do that it doesn't change the output in any…
-
Also covered reported alongside — the timeline has no entry for these yet
-
first by Interesting Engineering, 67d ago · also RuntimeWire, Silicon Republic, The Hans India, NDTV
4 more headlines
- Anthropic's invisible Claude watermark draws scrutiny over closed detection RuntimeWire · 67d ago
- Anthropic pledges to watermark AI content in EU Silicon Republic · 67d ago
- Anthropic adds invisible watermarks to Claude AI content The Hans India · 67d ago
- Anthropic Introduces Invisible Watermarks To Identify AI Content NDTV · 67d ago
-
first by Plagiarism Today, 64d ago · also The Hindu, HN Frontpage, Anthropic
2 more headlines
- Anthropic watermarking Claude AI content | Explained The Hindu · 64d ago
- How Claude's text watermarking works HN Frontpage · 64d ago
-
first by Android Authority, 67d ago · also Inc, The Verge
2 more headlines
-
first by Mastodon, 67d ago · also Business Insider, Yahoo Tech
2 more headlines
-
first by Mastodon, 62d ago · also Futurism
-
first by Forbes Business, 62d ago · also Forbes
-
first by PCMag, 63d ago · also PCMag Australia
-
first by Forbes Business, 67d ago · also Forbes
-
first by Daily Wire, 67d ago · also The Daily Wire
-
2 outlets AI cheating to be exposed by watermarks
first by Telegraph, 67d ago · also The Telegraph
and 24 smaller pieces
What people are saying 14 voices from 5 sites · best of 394 · verbatim
- Will false positives force people to alter genuinely human-written text that gets flagged as AI-generated?
- Will AI providers converge on a shared watermarking standard, and would that make removal harder or easier?
- Aug 17
-
I commented it last time the post about Claude watermarking went viral and I'm going to say the same thing again:"I've been working in the media and model IP space for quite many years. What this article misses a bit is the threat model for watermarking in general. Watermarking and fingerprinting have inherently weak security guarantees -- they…
-
L
If you're worried about Anthropic's Claude AI watermarking text it creates or processes ... THEN DON'T USE IT. I sure as hell don't, whether it's watermarked or not. And I don't use any of the other LLM AI chatbots either. Golly, I'm still able to write and code without AI. Inconceivable!
- Aug 15
-
T
Finally a good explanation of the text watermarking Anthropic will apply to Claude-generated content: https://www. anthropic.com/news/claude-text -watermark . They even link to the SynthID paper their approach is based upon, and have a real-world example so even lay persons should be able to understand the technique. I don't blame them for using…
- Aug 14
-
Good. If it doesn't catch the person that paraphrased, if it doesn't catch the person that used the AI as a tool and then put their on thoughts in their own result, but if it catches "You" who simply copied and pasted an AI response, then that watermark is doing exactly what it's supposed to do. If you hand in your dissertation, and your advisor…
- Aug 13
-
Suppose I gave you a series of 1000 coin flips. I tell you they were generated by a fair coin. You’re suspicious, you think I hid a watermark in them. But you look at the sequence and 485 are heads, close enough to 50%. You look at the correlation between a coin flip and the next and it’s 0.0433. Close to zero. You do a bunch more stats and…
-
Open weight models don't do this and erasing the statistical watermark is not hard and simply requires reprocessing the output with one of those models. It will likely become a common workflow for those who are worried that takes seconds. Once tools are available to identify the watermark, it will become even easier to remove. So this really only…
-
From: https://support.claude.com/en/articles/16266773-how-claude-m...> We'll support users and other third parties to detect Claude’s marksThis is the part that I fear most. What happens with false positives? Will I need to change arbitrary parts of my text if they are flagged as AI generated? Will I be forced to add mistakes?And then, since LLMs…
- Aug 11
-
Personally I wouldn’t have a problem if providing watermark removal was just made illegal, it’s not like AI forgery has all these legitimate use cases. You’d still have the issue of local models, but as you said, the important part is for it to be too much of a hassle for the average user. Also, worth noting that part of the point of steganography…
-
To a point this technology has existed for a while. There’s tools in the software industry to try and fingerprint source code to find open source code that is not properly attributed, or outright license incompatible. Those tools need to handle not only simple copy/paste but also someone just going in and changing variable/function names. It’s…
-
this is interesting; presumably apart from EU compliance this is also about preventing model collapse in the longer-term, so they *should* be incentivized to do it—and for book-length texts, it certainly ought to work
-
If LLM watermarks became industry standard, it'd be very interesting. I’ve seen people saying scammers would retype the document but… I dunno. I still regularly see chatbot prompts left in emails or student assignments. Most people are very lazy.
-
T
Anthropic is making its AI-generated content easier to identify, in compliance with EU rules.
-
Anthropic says new Claude models will embed invisible watermarks in all generated text, everywhere Claude is offered. The watermark is part of the text, it isn't metadata: “it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing.” [image]
-
If I read Anthropic's new watermarking policy correctly, not only will Claude watermark text it writes as AI (fair enough), if I upload text *I* write and ask Claude to copy edit, Claude will now watermark my (human written, AI-copy edited) text as AI, which seems ridiculous... [image]