Anthropic rolls out invisible AI text watermarks to comply with EU AI Act
ByPulseAugur Editorial·[105 sources]·
Anthropic has begun implementing invisible watermarks in its Claude models to identify AI-generated text and images, complying with the EU AI Act. This technology, developed in collaboration with researchers like Scott Aaronson, embeds subtle patterns in the model's token choices, making it detectable by a specific key without affecting output quality. While Google has used a similar system for Gemini, and OpenAI plans to follow suit, the effectiveness of these watermarks is debated, with tools already emerging to strip them, raising questions about attribution and the future of AI content identification.
AI
IMPACT
Sets a precedent for AI content attribution and transparency, potentially impacting how AI-generated content is consumed and regulated.
RANK_REASON
Product launch of watermarking feature by a major AI lab in response to regulatory requirements.
arXiv:2506.22343v2 Announce Type: replace-cross Abstract: Text watermarks in large language models (LLMs) are an increasingly important tool for detecting synthetic text and distinguishing human-written content from LLM-generated text. While most existing studies focus on determi…
Ahead of AI (Sebastian Raschka)
TIER_1English(EN)·Sebastian Raschka, PhD·
Anthropic sets out how it will mark Claude's text as the new transparency obligations take effect, while experts reject claims that watermarking degrades AI writing.
<p><a href="https://scottaaronson.blog/?p=6823">Scott Aaronson,</a> while working at OpenAI, largely solved AI text watermarking together with Hendrik Kirchner.</p> <p>Here is how his solution works, <a href="https://x.com/tenobrus/status/2089080656026566806">or see Tenobrus’s ve…
AI watermarks can be stripped with a free tool in minutes. But regulators and AI labs still expect them to stick around. Will they? Here's what that means for you today.
Hacker News — AI stories ≥50 points
TIER_1(AF)·padolsey·
Anthropic has pledged to start marking Claude-generated text and images with machine-readable data, in an effort to comply with European rules for AI transparency. "Generated text will carry embedded watermarks, and generated files will include digitally signed provenance metadat…
A little walkthrough explaining how Claude's new text watermarking works:
- How watermarking relates to the regular LLM sampling process
- Whether watermarking makes text "worse"
- How to remove watermarks
- How new text is checked for watermarks
magazine.sebastianraschka.com/p…
<div class="medium-feed-item"><p class="medium-feed-snippet">AI generated writing is becoming complex to differentiate from human-written content. However it sparks a question ”if AI generates…</p><p class="medium-feed-link"><a href="https://medium.com/@priyasengar2…
Users are developing tools to remove Claude AI text watermarks, sparking debate over AI content detection and attribution. The race to strip watermarks raises questions about the future of AI-generated content identification. https:// gizmodo.com/people-are-rushing -to-find-ways-…
<h2> What Happened </h2> <p><a href="https://dev.to/go/claude">Claude</a> now appends a visible tag to every generated response. The tag, e.g., <em>“Generated by Claude”</em>, appears inline with the text. Anthropic added the feature to make AI‑generated content obvious and to me…
I do wonder how "invisible" a watermark in # AI generated text actually is: https://www. computerworld.com/article/4208 149/anthropic-will-begin-watermarking-ai-generated-content.html # ArtificialIntelligence
I do wonder how "invisible" a watermark in # AI generated text actually is: https://www. computerworld.com/article/4208 149/anthropic-will-begin-watermarking-ai-generated-content.html # ArtificialIntelligence
<p>Anthropic says it will change way chatbot makes small, random choices, to comply with EU regulation</p><p>The world is familiar by now with the usual tropes of machine-generated text: overuse of the word “delve”, an excess of em dashes, and the <a href="https://www.theguardian…
Can a watermark settle whether a document was written by AI? A visual explainer walks the mechanism in five steps, with dice you can roll to watch a secret key tilt the model's word choice. But Anthropic's implementation can only detect texts written by its models, and can't work…
<p>Watermarks prove an AI wrote the content. Almost nobody is proving what the agent decided, or that the record was not changed afterward. That second half is where the money, the access, and the liability live.</p> <p>This month Anthropic started watermarking everything Claude …
# AI # text # watermarking works by embedding a pattern in the choices between words made by a language model. A secret key subtly influences these choices, creating a detectable pattern without altering the text’s meaning. This # watermark can be verified by the key-holder, but …
<p>Something changed on August 2, 2026. Every new Claude model started embedding an invisible mark into the text it generates. You can't see it, strip it with backspace, or find it with Ctrl+F. But it's there — baked into which words were chosen.</p> <p>Google's been doing the sa…
The chatter about # Claude & # watermarks likely precedes the next wave of # AI monetization: watermark detectors, tools promising to evade them, 3rd-party “verification” gimmicks, etc. You can bypass the whole ecosystem, of course, by not using Claude if you intend to pass its w…
Will this invisible plaintext # AI “watermarking”work? https:// support.claude.com/en/articles /16266773-how-claude-marks-ai-generated-content # Claude
Claude will apply invisible watermarks to AI text and images https://www.theverge.com/ai-artificial-intelligence/977823/anthropic-claude-ai-watermarks-c2pa-text-images # AI # Privacy # Tech
Thoughts on AI watermarking Claude is now watermarking its text outputs using a statistical token-choice watermark. According to what the AI told me, when generating the next token, it often has a choice between options that have similar meaning. It applies a statistical bias for…
🤖 Will Chinese Open Source Agree to EU Watermarking? I wonder if people are thinking and worried about this yet? Anthopic, OpenAI and the western AI labs have agreed to watermark AI outputs. Some of us want free and open and untracked and un-modified... 📰 Source: Artificial Intel…
AI Watermarks Remover - Strip multi-vendor AI provenance marks and bypass "Generated with AI" detections with ease https:// github.com/guillaumemeyer/wate rmarks-remover # aiwatermarksremover # foss # opensource # github # AI
AI firms are watermarking generated text – here’s why it won’t work. AI companies have begun embedding watermarks in the output of their models to improve transparency and help crack down on misinformation, disinformation and cheating on homework, but there are limitations to thi…
More AI models will be adding watermarks soon (like SynthID). I guess partially due to the EU AI act. But I was thinking, might it be to prevent the AI models from eating their digital poo as input? # AI # DAIrrhea
# AI watermarking is in the news due to Everyone-But-Elon signed up to the EU watermarking declaration (is this like a pinky promise?) to use SynthID. It works by measurable token-biasing to give a degree of confidence but not certainty that it has been merely touched by AI. http…
How AI text watermarking works: a visual guide "A gentle visual guide to how statistical watermarks hide inside AI-generated text, and what erases them." Link: https:// declaude.org/watermarking/ # linkdump # ai # watermark
RE: https:// helvede.net/@jwcph/11705351145 8535906 Yeah, about that "AI watermarking" thing (that is actually more like fingerprinting but words amirite...) # AI
More on watermarking # AI generated text: https://www. extremetech.com/internet/anthr opic-to-add-invisible-watermarks-to-claude-generated-text # ArtificialIntelligence
More on watermarking # AI generated text: https://www. extremetech.com/internet/anthr opic-to-add-invisible-watermarks-to-claude-generated-text # ArtificialIntelligence
Someone built a free tool to scrub # AI watermarks from # OpenAI , # Gemini -generated text and files https:// tech.yahoo.com/ai/deals/articl es/someone-built-free-tool-scrub-041325693.html
AI Watermarking and Transparency Laws. Claude is the first major LLM to legally watermark text generated by, or edited with, their tools. Open AI (ChatGPT) is to follow suit in the coming weeks, amongst most other providers: https://www. lexacademic.com/blog/the-stamp -of-shame-w…
The whole "AI" watermarking thing is revealing. That's understating it. Background: one of the LLM frauds announced they were going to make their model watermark its output text so that it would be detectable by an "AI"-detection tool. Many of the others have indicated they're go…
So # ai text watermarking is hidden in token selection, not visibly added to text. - Kirchenbauer: make green tokens more likely than red using a secret key (bias the dice) - Aaronson: sample from the model's distribution while controlling the randomness with a secret key (contro…
Jak odpowiedzi AI będą znakowane w sposób niewidzialny dla ludzi? Niebawem wszystkie znane narzędzia AI zaczną dodawać w niewidoczne dla ludzi znaki do generowanych przez siebie odpowiedzi (zarówno tekstu, obrazów jak i kodu). Te tzw. znaki wodne (ang. “watermark”) to nie będą ni…
New article: Digital Signatures and Watermarks — Defending Authenticity in the Era of Generative AI. As generative AI makes convincing fakes easier, robust authenticity tools become essential. This piece explains how digital signatures and watermarking work, where they fit in com…
Let's be honest, Claude's watermark will work like any other AI detection system. Great if you speak US English natively; shitty if you don't. # AI # detection # watermark https:// techcrunch.com/2026/08/15/anth ropic-shares-more-details-about-how-claudes-new-watermarks-will-work…
Claude is now watermarking writing. The watermarking system is called SythID-Text. There's a law in the European Union that mandates watermarks (EU AI Act Article 50). Other companies (OpenAI, Google, etc) will follow suit. https://www. youtube.com/watch?v=Ucb8lk5TlhA https://www…
Claude watermarking for EU AI Act compliance is getting very close. This is a good summary of what that means. tl;dr: Sub-patterns of largely fungible token choice will skewed in non-distinguishable ways that will, based on volume of impact, indicate likelihood of Claude’s involv…
🤖✨ Ah, watermarking! The cutting-edge # technology where # AI gently scribbles "I wrote this" in # invisible # ink 🙄. It's like Claude is leaving a secret handshake in every sentence, all while not affecting a single thing. Because nothing screams # compliance with # bureaucratic…
Ah, the "deep" mysteries of # AI text watermarking! A gripping tale of invisible ink for the digital age—minus the ink or the intrigue. 🔍 Spoiler alert: the real magic trick is convincing you there's a mystery to solve. 🤡📜 https:// declaude.org/watermarking/ # Watermarking # Digi…
🌱 Spotted early: guillaumemeyer/watermarks-remover — an open-source project that strips AI provenance marks from text, metadata, and C2PA across multiple vendors. Interesting take on the watermarking debate. See how it works: https:// olud.ai/latest.html # OpenSource # AI # GitHu…
A GitHub tool for removing AI watermarks gained 4,100+ stars in two days. It verifiably strips metadata and hidden characters, but its central claim—defeating Claude's text watermark—cannot be tested. Anthropic has not published a detector or technical spec. https://www. implicat…
All major AI companies now add a digital watermark to all content generated by their AI models Even AI-generated text is watermarked, in a way that is preserved even some level of editing. But how is that possible? We try to explain the basics of the LLM's text watermarking techn…
Kaikki suurimmat tekoälypalvelut lisäävät nyt "vesileiman" tuottamaansa sisältöön - myös tekstiin Ja tekstissä oleva tunniste säilyy osana tekstiä, vaikka tekstin kopioisi ja liittäisi muualle - ja jopa senkin jälkeen, kun tekstiä muokkaa. Vastaava käytäntö on käytössä nyt sekä O…
Delving into the world of Claude Watermarks, we uncover how AI-assisted work can be scrutinized. It shines a light on hidden AI writing, but also brings up challenging issues regarding transparency, privacy and fairness in detection. Are we ready for this new frontier? # AI # Cla…
Claude mette una firma invisibile ai suoi testi: cosa cambia davvero con il watermark dell’IA 📌 Link all'articolo : https://www. redhotcyber.com/post/claude-me tte-una-firma-invisibile-ai-suoi-testi-cosa-cambia-davvero-con-il-watermark-dellia/ Inva Malaj # redhotcyber # cybersecu…
Claude is adding watermarks to generated content Due to EU transparency laws around AI-generated content, future models will add embedded watermarks and digitally signed provenance metadata. https:// janefriedman.com/claude-is-add ing-watermarks-to-generated-content/ # AI
Как работают текстовые ИИ-вотермарки и как их обходить Наверное уже многие знают, что в силу недавно вступивших в ЕС законов, регулирующих использование ИИ, провайдеры LLM обязаны помечать ИИ-сгенерированные тексты. Та же Claude недавно обновила справочную информацию о работе сво…
🔍 Claude's new invisible watermark is live worldwide—driven by EU AI Act rules. If you run a small business using AI, here's what actually changes 👇 https:// signaldigital.net/2026/08/12/w hat-does-anthropic-watermarking-mean-for-small-business-in-the-eu/ # AIAct # SmallBusiness …
<h2> The Copy-Paste Conundrum: When AI Text Becomes 'Yours' </h2> <p>The cursor blinks. It’s a simple act, one performed millions of times a day across the globe: highlighting a block of text, pressing control-C, then control-V. A paragraph for a report, a few lines of code, an e…
dev.to — LLM tag
TIER_1English(EN)·Sylwia Laskowska·
<p>For the past few hours, the whole world, or at least my LinkedIn feed, has been talking about one thing: Anthropic signed the <strong>EU AI Act’s Code of Practice on Transparency of AI-Generated Content</strong>.</p> <p>Is this the end of generating funny cat storie…
🤖 Claude va désormais ajouter un filigrane invisible à ses textes générés par IA 👉 https://www. justgeek.fr/claude-filigrane-i nvisible-textes-156741/ # Claude # Anthropic # Filigrane # Watermark # Texte # IA # AI
Claude marquera le contenu généré par IA pour se conformer à l'AI Act : watermarks invisibles dans le texte, métadonnées C2PA signées dans les fichiers, dès août 2026 pour les nouveaux modèles UE. ⬇️ https:// support.claude.com/en/articles /16266773-how-claude-marks-ai-generated-…
Anyone using AI to write for them may want to pay close attention: your work is about to be watermarked, and systems will in fact know. Human may not. https://www. pickr.com.au/news/2026/ai-wate rmarking-means-text-can-be-checked # news # online # ai # anthropic # claudeai # gemi…
Someone built a free tool to scrub #AI watermarks from #OpenAI , #Gemini-generated text and files tech.yahoo.com/ai/deals/art... Someone built a free tool to s...
Anthropic erklärt, wie unsichtbare Text-Wasserzeichen in Claude funktionieren: eine Variante von Googles Open-Source-SynthID-Text, gedacht zur Einhaltung der EU-Transparenzpflichten. Kritiker bezweifeln aber, dass die Wortwahl dabei wirklich unbeeinflusst bleibt. https://www. the…
Claude is watermarking every response — here's what that means if you use AI for writing Claude now secretly watermarks every AI-generated response. Here's how AI watermarking works, whether it can be removed and what it means for power users. https://www. tomsguide.com/ai/claude…
guillaumemeyer/watermarks-remover strips AI provenance marks like Unicode text hygiene and C2PA metadata — it hit 1.7k★ fast because it targets the very signature that identifies content as AI-made. https:// olud.ai/latest.html # OpenSource # AI # GitHub
<table> <tr><td> <a href="https://www.reddit.com/r/ClaudeAI/comments/1vlidn0/claude_will_now_include_invisible_marks_to_show_a/"> <img alt="Claude will now include invisible marks to show a text was made with AI" src="https://external-preview.redd.it/X9YTNUZrN-YoLXmRMcG4zN6NEZ58j…
<!-- SC_OFF --><div class="md"><p>Anthropic's recent announcement about adding invisible watermarks to generated text got me curious about how a watermark can exist in plain text without relying on metadata or hidden characters.</p> <p>I ended up reading about a few of the approa…
<!-- SC_OFF --><div class="md"><p>I saw that Anthropic is apparently planning to embed invisible watermarks into Claude-generated text, and this part caught my attention:</p> <blockquote> </blockquote> <p>Does anyone here know how something like this actually works technically?</…