Anthropic begins invisible watermarking for Claude text and images to meet EU AI rules

Claude / AnthropicContinue reading on Medium »

#Anthropic #Claude #AIWatermarking #AITransparency #GenerativeAI #AISafety #AI #ArtificialIntelligence #Technology #RubbletagContinue reading on Medium »

Anthropic announced on August 14 that Claude now embeds a hidden watermark in its text outputs — invisible to readers, undetectable by…Continue reading on Medium »

Anthropic announced last week it would include invisible watermarks in AI-generated content to comply with new EU rules. Within hours, overrides were being touted online.

Anthropic is adding invisible watermarks to everything Claude produces — and to everything Claude merely edits. Under the EU AI Act’s Article 50, providers of AI systems that generate text must mark their outputs so they can be detected as machine-made. Anthropic’s implementation, announced in August 2026, applies that mark at the token-selection level: every time Claude chooses a word, the choice is nudged to encode a hidden signal. The consequence is that the technique cannot distinguish between text Claude authored from scratch and text a user wrote themselves but asked Claude to polish. If Claude touched it, it is marked. The answer-first version: Anthropic is watermarking all Claude output, worldwide, by biasing which near-equivalent words the model selects. The mark is invisible, cannot be opted out of, and applies to edited text as well as generated text. If you use Claude to proofread your own writing, your words now carry a signal that says “processed by AI” — and you have no way to see it, remove it, or check how much of your text it affects. What Anthropic announced In an updated support article, Anthropic confirmed it has signed the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content. The article describes how the company plans to comply: by embedding watermarks in text, and signed provenance metadata in generated files such as PDFs, images and SVGs. For text, the key passage reads: “When a supported Claude model generates text, 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.” Notably, the watermark applies worldwide, not only in the EU. Anthropic states that “marking will apply to output from supported models wherever Claude is offered” — meaning users in every jurisdiction receive the same treatment, regardless of whether their local law requires it. The company has not explained why it chose global application over a regional one, though a per-region carve-out would obviously weaken the signal’s utility. How the watermark works When a language model generates text, it does not write words; it samples tokens. At each step, the model produces a probability distribution over possible next tokens — the word “tropical” might have a 12% chance, “exotic” 11%, “fresh” 9% — and selects one. Normally this selection involves randomness: run the same prompt twice and you may get different phrasing. Statistical text watermarking works by replacing that randomness with a deterministic, key-dependent selection. Among tokens that are near-equally probable, the model consistently picks the one that encodes a bit of the watermark signal. Run enough tokens through this process and the pattern becomes detectable by anyone holding the key. The technique is real cryptography, not security theatre. It has been studied in academic literature and its properties are well understood. As one commenter on Hacker News noted, “the very fact that there is generally no ‘best next token’ with 100% certainty is precisely why the trick works.” The output is provably drawn from the same distribution the model would have sampled anyway — the watermark selects which equally-good token, not a worse one. The proofreading problem Here is where the maths and the lived experience diverge. When you ask Claude to proofread or lightly edit your own text, the model does not return your text with red-pen annotations. It regenerates the entire response token by token, making a selection at every position. Even where the “right” token is obviously the same word you already wrote, the model is still choosing it from a probability distribution, and the watermark still biases that choice. The output may read identically to what you wrote, but it now carries a statistical signature. Anthropic’s own phrasing is careful: a detected mark indicates the content “may have been processed by Claude.” Note the word “processed”, not “generated.” This is honest — it accurately describes what the watermark detects. But downstream systems that check for AI-generated content are unlikely to make the same distinction. A university plagiarism detector, a publisher’s AI-content filter, or a job applicant screening tool that flags watermarked text will read “AI-generated” whether the human wrote 90% of the words or none of them. A watermark that cannot tell the difference between “Claude wrote this” and “you wrote this but Claude was in the room” is not labelling authorship. It is labelling contact. What the EU AI Act actually requires Article 50 of the EU AI Act requires providers of AI systems that generate synthetic audio, image, video or text content to ensure their outputs are “marked in a machine-readable format and detectable as artificially generated or manipulated.” The regulation targets synthetic content — content the AI created — not content a human created and an AI relayed or lightly edited. But the watermarking technique, by its nature, does not make that distinction. Everything Claude outputs gets marked, because everything Claude outputs passes through the same token-selection process. There is a reasonable argument that Anthropic is over-complying: applying the mark globally and to all processed text, rather than limiting it to text Claude substantively generated. There is also a reasonable argument that over-compliance is the safer legal position, since drawing the line between “Claude wrote this” and “Claude edited this” is exactly the sort of judgement call a regulator might second-guess. Anthropic chose the belt-and-braces approach, and the users who treat Claude as a writing tool rather than a writing replacement bear the cost. Why this matters more than it looks Several consequences follow, none of them hypothetical. First, anyone who uses Claude as a writing assistant — to proofread emails, sharpen cover letters, edit blog posts — now has their own writing flagged as AI-processed. In any context where AI-generated content is penalised (academic submissions, journalism, certain job applications), this creates a new risk that did not exist before the watermark. Second, the watermark is invisible and there is no opt-out. Anthropic’s terms require users not to remove it, but since users cannot see it, they cannot know how much of their text it affects or whether a particular passage still carries the signal after editing. The detection tools are not yet public either — so users are asked to accept a mark they cannot see, on terms they cannot verify, with detection they cannot access. Third, the watermark persists. Anthropic says it “will travel with the text when it’s copied and pasted elsewhere, and may persist through some editing.” A paragraph you asked Claude to tidy up in January may still carry its mark when a client or employer runs it through a detector in June. For anyone whose relationship with their own data already feels tenuous, this is another piece of control quietly transferred to the provider. What users can do The practical advice is unglamorous: Ask for advice, not rewrites. If you need text to remain demonstrably yours, do not have Claude return the text directly. Ask it to critique a paragraph, suggest improvements or flag errors — then make the changes yourself. This is the only way to avoid the mark. Retype, do not paste. Copy Claude’s suggestions and retype them in your own document. The watermark lives in the token stream Claude generated; retyped text carries no signal. Know the limits for code. The watermark applies to generated code tokens too, though highly constrained output (structured data, code with strict syntax) leaves less room for near-equivalent token selection and therefore carries a weaker signal. How much weaker is unknown until Anthropic publishes its detection documentation. Watch for the detector. Anthropic says detection tools are coming. When they arrive, test your own edited text against them — that will tell you more about the watermark’s practical reach than any amount of speculation. What this comes down to is a question about whose words they are. Anthropic’s safety and compliance obligations are real, and the EU AI Act’s transparency goals are defensible. But a watermark that cannot distinguish generation from editing treats every Claude user’s output as AI content by default. For anyone who thought they were using a tool to improve their own writing, the tool is now asserting a claim over it — invisibly, irrevocably, and worldwide. Originally published at theaidownside.com — evidence-first reporting on the costs and trade-offs behind AI products.

Attempt to mark out AI-generated text has led to intense criticism – and explanation has only made it stronger

Anthropic explains how Claude's new text watermarking works, why it's rolling out, and what it can and can't prove.

Anthropic has explained the technology behind Claude’s upcoming text watermarking system, which will use word-selection patterns to indicate whether the AI contributed to a passage. The company says the method will not alter response quality, add hidden characters or increase costs. Detection will be less reliable for short, factual or lightly edited text, while code will generally carry less of the watermark. Anthropic is also preparing a detection API and will apply C2PA metadata to supported files created or processed by Claude.

Anthropic has clarified how it's planning to apply invisible watermarks to Claude-generated text in order to comply with Europe's AI transparency rules. On Friday, Anthropic announced that Claude's text marking system is "a version of the SynthID-Text approach" - an open-source watermarking technology developed by Google DeepMind that creates detectable patterns using wording probabilities. This […]

Anthropic has detailed how its AI model Claude embeds machine-readable watermarks into generated text, as part of the company's compliance with Article 50 of the European Union's AI Act. The transparency provisions under this article took effect in August, requiring providers of generative AI systems to mark outputs as artificially generated wherever technically feasible.How the watermark works?According to Anthropic, the watermark does not rely on hidden characters or metadata tags, since plain text carries no space for such additions. Instead, the company said the system embeds an imperceptible statistical pattern into the model's word choices during generation. When Claude has multiple equally natural ways to complete a sentence, the selection process uses a secret mathematical key rather than standard randomness. Anyone holding Anthropic's corresponding detector key can then check whether a piece of text matches the pattern Claude would have produced under that key.Anthropic has said the mark does not alter the meaning, quality, or readability of a response, and that it can persist through copying, pasting, and some editing.Where the watermark appliesThe marking is being applied globally rather than only to users in the EU, and covers Claude models launched on or after August 2, across the Claude apps, API, Claude Code, Cowork, and Tag, as well as deployments through AWS, Google Cloud, and Microsoft Foundry. Anthropic has said it is working to extend the capability to older models, without a confirmed timeline. Generated image files in formats such as .svg, .png, and .jpg carry separate signed provenance metadata based on the C2PA standard, used to indicate how a file was created and whether it has since been altered.Limitations acknowledged by AnthropicAnthropic has stated that a detected mark indicates Claude may have processed a piece of content, not that it necessarily authored it, since the model is also used to translate, summarise, proofread, or edit text originally written by a person. The company has also acknowledged that marks may not register on very short passages or narrowly factual text, since there are fewer open word-choice decisions for the pattern to embed itself in. Anthropic has not published the technical implementation of the watermark, and independent researchers have previously demonstrated that watermarking schemes of this kind can, in some cases, be spoofed or removed through heavy paraphrasing, translation, or rewriting with another model.The requirement stems from the EU AI Act's push to make synthetic content identifiable, with non-compliance carrying fines of up to €15 million or 3 percent of a company's global annual turnover, whichever is higher. Anthropic has signed the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content as both a model provider and a system provider. Companies deploying Claude within their own products have been told to separately assess their own obligations under Article 50.

NPR's Michel Martin speaks with Fortune Magazine AI reporter Beatrice Nolan about the invisible watermark embedded in anything new models of Anthropic's AI assistant Claude processes.

Anthropic noted that Claude’s text watermark was a version of the SynthID-Text approach published by Google DeepMind in a paper two years ago

There are situations where Claude's watermarking solution might not be effective.

The feature is designed to comply with updated European transparency requirements. But some users fear it will flag even lightly edited AI-assisted text like a modern-day ‘scarlet letter.’

This week Anthropic announced its Claude chatbot will watermark the text it generates, reports the blog Futurism. "It works by making subtle changes in the AI's word choices across the text it generates, which are supposed to be imperceptible to a human but, in aggregate, form a pattern that is detectable with the tool." Anthropic said it was implementing the watermark system in response to the European Union's landmark AI Act passed in 2024, which requires that AI companies mark content that's been generated or edited by their systems. TechCrunch notes that other companies including Google, Meta, Microsoft, OpenAI, and Synthesia have committed to adhering to the EU's code. But "The news about Claude watermarking kicked off a firestorm on X," reports Forbes, "with users panicking that AI-assisted writing will now bear a kind of permanent 'scarlet letter.'" (Forbes wonders if media companies are normalizing AI "while at the same time stigmatizing the output.") But there's another issue. Anthropic said the watermark indicates Claude processed text, Business Insider points out, "not that it was necessarily the original author. The company said marks can remain after Claude proofreads, translates, summarizes, or otherwise edits content... a watermark shows that Claude processed text, not necessarily that it wrote it." In the days since, "Dozens of people have posted on X since Monday that they had canceled their Claude subscriptions, citing the watermark." Vladislav Rajtmajer, a freelance developer in the Czech Republic, told Business Insider he canceled his Claude Max subscription on Tuesday, citing the watermark as the main reason. He said he uses AI for code reviews and translations, and worried an AI label on code shipped to clients could raise questions about authorship or trigger contract penalties... Richard Echols, an AI consultant in Georgia, also said he canceled his Claude Max subscription on Wednesday and that the watermark was a major factor in his decision. He said he uses Claude to create documentation for businesses and worried the mark could appear when he had written the underlying material himself and used the chatbot only for edits. "Even if you edit your own work that you wrote, they add the watermark anyway," Echols said... Anthropic isn't the only AI lab to use watermarks: Google uses its SynthID technology to watermark AI-generated content, and OpenAI uses SynthID for supported images and audio. Elon Musk's social media platform X also adds a "Made with AI" tag on content it determines to be AI-generated or manipulated. Long-time Slashdot reader kmleon (also a tech consultant and RPG researcher) experienced another issue. He'd built some software using their own custom-built AI tool, only asking Claude to do some testing. "Little did I know that Claude would sometimes decide, since it was involved on the edges... to take credit for all of my work, claiming some parts were either "100%" or "Co-Authored/Created" by Claude.Ai / Claude Code!" I'm sure the good folks at Anthropic will eventually tweak Claude enough to, hopefully, not make these mistakes in the future.... I only became aware of it today, because I recently setup scanners to catch Claude incorrectly watermarking any content it touches for various projects I'm working on. I found that it has (apparently as far back as Sonnet 4.6, but as recently as today, August 14th, 2026) sometimes been mistakenly over-zealous in taking credit for 100% creating or co-authoring docs, code, images, videos, etc. I've only, so far, found about a dozen instances spanning the past year, and most of them appear to have been injected in the last few weeks, so it is not a common issue (yet). On the bright side? Maybe watermarks can help AI stop training on AI-generated text. Although Forbes also reports that Google "announced on Friday that it will now let users remove the visible watermark from AI-generated images, videos and music created with several of its models." Read more of this story at Slashdot.

Anthropic now embeds invisible watermarks in Claude-generated text. Here’s how Claude’s AI watermark works, where it applies, what it can…Continue reading on The Tech Trek by Tech Chick‍ »

Claude's watermarking is invisible to the naked eye, but is created via a select pattern in its responses, as per Anthropic.

Anthropic says it will add invisible AI watermark to show when you use Claude but it is okay

It could soon become easier to identify AI-generated content, even if it's not the usual "It's Not X, it's Y" type of post you'd come across on LinkedIn and other socials. [...]

Anthropic addresses concerns about its new watermark, which will affect Claude users globally. It said the watermark is to comply with the EU AI Act.

Anthropic is weaving an invisible, machine-readable watermark into every word its newest Claude models write—and it hasn't said how.

While many regulators want AI content to be more clearly labelled, Claude users are worried about the risks of their written work being flagged as AI

Anthropic announced they are turning on watermarking for their Claude AI-generated outputs. This is significant. An AI Insider analysis and scoop.

Anthropic, a leading AI company, is introducing watermarking for outputs generated by its Claude chatbot, enhancing transparency and complying with the European Union's AI Act that necessitates identifying AI content. Despite some user dissatisfaction with the digital watermarking, Anthropic is also setting auto mode as the new default for its Claude Code platform, striving to improve safety and operational efficiency in programmers' tasks.

The move is part of Anthropic's commitments under the European Union's AI Act, which includes transparency rules that kicked in early this month.

Following Anthropic's confirmation that all text generated by its new Claude models will carry an invisible watermark in order to identify that the text has been generated by AI. Read more about this measure at: https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content Today, developer Guillaume Meyer published "watermarks-remover" on GitHub: an open-source project that cleans those signals generated by LLMs, such as Claude, Gemini, OpenAI and others, removing invisible Unicode characters, C2PA metadata and more. 🔗 Repository link: https://github.com/guillaumemeyer/watermarks-remover

This is meant to make content generated by AI easier to detect.

Anthropic has confirmed a global plan to mark content generated by Claude with embedded text watermarks and signed provenance metadata for files. The initiative is designed to support transparency obligations under the European Union's AI Act, while giving users and third parties ways to detect Claude's marks. Crucially, Anthropic says a detection result is not definitive proof that Claude authored the original content, and a missing mark does not rule out AI involvement. The policy begins with Claude models launched on or after August 2, 2026 in the EU. Anthropic says it is working to bring older models into compliance during a transition period. The company plans to apply the approach worldwide wherever Claude is offered, including supported deployments through AWS, Google Cloud, and Microsoft Foundry. According to Anthropic's official guidance on how Claude marks AI-generated content, the program covers supported models used through the Claude Platform API, Claude, Claude Code, Claude Cowork, and Claude Tag. That scope makes the change relevant not only to direct Claude users, but also to businesses building Claude into internal tools, publishing workflows, and customer-facing applications. How Claude's content marking works Anthropic describes two distinct mechanisms. One applies to generated text, while the other provides file-level provenance information. They serve related transparency goals, but they are not interchangeable. Marking mechanism What Anthropic says it applies to Purpose Embedded watermark Generated text Provide a machine-readable signal that text may have been processed by Claude Signed provenance metadata Generated files Attach provenance information using the C2PA standard The text watermark is intended to be embedded in Claude output rather than displayed as an ordinary label for a reader. The provenance metadata uses C2PA, a standard for recording content provenance. In practice, these are separate signals: a text-based asset may be assessed for a watermark, while a generated file can carry signed metadata where the relevant format and platform support it. Anthropic's wording matters because the company does not present either mechanism as an infallible authorship test. A detected watermark indicates that content may have been processed by Claude. It does not establish that Claude created every part of the final work, that a user did not substantially edit it, or that Claude was the sole source of the original material. Likewise, the absence of a detectable mark is not evidence that content was written without AI assistance or was never processed by Claude. Marks may not be available for every format or platform at all times, and a final asset can pass through editing, conversion, or distribution workflows that affect the available provenance signals. Detection tools are planned, not yet detailed Anthropic says it will provide mechanisms for users and third parties to detect Claude marks, with technical details to follow in future documentation. That commitment is significant for platforms, publishers, and enterprise teams that may need to assess content received from outside their organization. For now, companies should avoid building policies around assumptions about a particular detector's interface, coverage, or reliability. The confirmed policy establishes the direction of travel, but the practical implementation details for detection remain to be published. Why the EU timeline matters Anthropic links the rollout to Article 50 of the EU AI Act's Code of Practice on Transparency of AI-Generated Content. The provision concerns labeling and detection of AI-generated content, placing transparency obligations on relevant AI providers and deployers. The initial date applies to newer models in the EU, followed by a transition period for older models. Yet Anthropic's stated worldwide scope means the change should not be treated as an EU-only product behavior. Businesses using Claude across multiple regions may ultimately encounter the same marking approach, subject to support for particular models, file types, platforms, and deployment surfaces. For organizations, the central compliance question is not simply whether a mark exists. It is whether content governance processes accurately document how AI was used, what human review occurred, and what disclosures may be required by the relevant platform, customer contract, or jurisdiction. What marketers and content teams should change Marketing teams should treat Claude marking as a provenance and workflow consideration, not as a verdict on whether content is acceptable to publish. AI-assisted work can be reviewed, revised, approved, and used within a responsible editorial process. But the ability to identify a Claude-related signal changes the expectations around record keeping and disclosure. Teams using Claude for copy, campaign assets, documents, or code should consider the following: Identify where supported Claude models enter the content production workflow, including API and cloud deployments. Document human review, edits, approval ownership, and the intended use of generated material. Check whether downstream publishing tools preserve or expose C2PA metadata for relevant file formats. Avoid representing watermark detection as conclusive authorship evidence in compliance, moderation, or dispute processes. Prepare for Anthropic's forthcoming detection documentation before making technical procurement or policy decisions. There is also a privacy and governance angle. Provenance metadata and detection systems can help support transparency, but organizations should understand what information their own tools retain, expose, or pass to third parties. Anthropic's announcement confirms the use of signed provenance metadata, not the specific data-handling policies of every platform that may process a marked file afterward. For businesses that rely on AI-assisted publishing at scale, this development reinforces the need to align marketing operations with legal, brand, and security teams. A watermark can be useful evidence within a broader chain of information, but it cannot substitute for an internal audit trail or editorial controls. As AI-generated content becomes easier to identify through machine-readable signals, brands need defensible processes for visibility, governance, and disclosure. Scalevise's AI consultancy helps organizations assess AI content workflows, define practical controls, and turn emerging platform changes into accountable operating practices. Request a consultation to map your Claude use cases and content governance priorities. Frequently Asked Questions When will Anthropic begin marking Claude-generated content? Anthropic says marking begins in the EU for Claude models launched on or after August 2, 2026. The company is working to add support for older models during a transition period. Which Claude products are covered by the watermarking plan? Anthropic says marks will apply to supported Claude models across the Claude Platform API, Claude, Claude Code, Claude Cowork, and Claude Tag, wherever Claude is offered worldwide. Does a detected Claude watermark prove Claude wrote the content? No. Anthropic says detection indicates content may have been processed by Claude, but it is not conclusive proof that Claude authored the original content. Does no watermark mean content was not generated or processed by AI? No. Anthropic states that the absence of a watermark does not prove content was not AI-generated or processed by Claude. What is C2PA metadata in Anthropic's plan? Anthropic says it will attach signed provenance metadata to generated files using the C2PA standard. It is separate from the embedded watermarking Anthropic plans for generated text. Conclusion Anthropic's plan makes content marking a built-in consideration for supported Claude deployments, beginning with newer EU models and extending globally over time. Text watermarks and C2PA metadata can improve transparency, but Anthropic's own limitations are equally important: neither detection nor its absence settles authorship. Businesses should prepare by strengthening review, documentation, and disclosure processes rather than treating a single technical signal as a complete compliance answer.

Anthropic has announced that new Claude models will hide an invisible watermark inside any generated text — here’s how that’s even possible, and how OpenAI and Google might have to follow suit.

The American AI company Anthropic announced it will add an invisible watermark to text produced by its Claude chatbot so as to comply with European Union regulations. But the company noted that this mark only means the text was processed by Claude, not that the chatbot actually generated it. The new feature unveiled on Monday in San Francisco will not be visible to humans and will remain in place even after text is cut-and-pasted. But the tool has limitations, Anthropic said. Detecting a Claude watermark means the text may have been processed by Claude but does not confirm the source, the company said in a statement. For example, it said, “Claude may not be the original author. People often use Claude to proofread, translate, summarise, or convert files. The output can carry a Claude mark even if the underlying ideas, text, or data originated from another source.” Conversely, a text that is very short or has been heavily edited may not carry a watermark. When Claude generates a supported file type like .jpg, it will attach signed provenance metadata in line with an open standard used to record information about content provenance, Anthropic said. It said only Claude models launched on or after August 2 will feature the new watermarks. Earlier ones will be upgraded. These changes are meant to comply with new EU regulations on artificial intelligence that say AI-generated content must feature machine-readable markings. Violations risk drawing stiff fines. Anthropic said it will use these watermarks all around the world, not just in Europe. Anthropic is among some 190 organisations that signed a voluntary EU code of good practice late last month, with others including Google, Meta, Microsoft, Mistral and OpenAI. Google has come up with SynthID, a non-public advanced digital watermarking and detection technology. It is pushing for industry giants like Apple, Nvidia and OpenAI to adopt this standard. Anthropic has not yet released a watermarking detection tool as called for in the EU code of good practices.

Anthropic has revealed how Claude will mark AI-generated content — and the approach is more sophisticated than visible labels or metadata tags. Claude will weave imperceptible watermarks directly into generated text, following the EU AI Act's Article 50(2) Code of Practice on transparency of AI-generated content. How It Works The watermarking operates at the model level — meaning it's applied directly in the text itself, not as metadata or a separate header. According to Anthropic's support documentation: Plain text output: Claude weaves an imperceptible watermark directly into the text. Users won't see it, and it doesn't change the meaning, quality, or readability of the response. Persistence: Because the watermark is part of the text, it travels with the text when copied and pasted elsewhere, and may survive some editing. Model-level application: The watermark is applied at the model level, meaning it's present regardless of which interface or application is used to access Claude. This is fundamentally different from approaches like OpenAI's, which previously explored metadata-based watermarking that could be stripped by copying text out of the original interface. Anthropic's approach embeds the watermark in the text itself — making it far more durable. The EU AI Act Connection Anthropic has signed the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content. This code requires providers of generative AI to mark AI-generated content in a machine-readable way that can detect manipulation or forgery. The commitment covers both: Generative AI models (like Claude itself) Generative AI systems (applications built on top of Claude) This means the watermarking applies not just to direct Claude interactions, but to any application that uses Claude's API to generate content. What This Means for Content Creators For developers and content creators using Claude, this has several implications: Transparency by default: Any content generated by Claude will carry the watermark, whether the creator wants it to or not. This is a feature for platforms fighting AI spam, but could be a complication for legitimate AI-assisted workflows. Copy-paste persistence: The watermark survives copy-paste operations, meaning AI-generated content can be identified even after being moved to a different platform or document. Editing resilience: The watermark "may persist through some editing" — suggesting it's not a simple flag that gets removed by changing a word. The exact resilience threshold isn't specified. API implications: Since the watermark is at the model level, it applies to API-generated content too. Applications using Claude's API will produce watermarked text without needing to implement watermarking themselves. The Detection Side The key question is: how do you detect the watermark? Anthropic's documentation doesn't detail the detection mechanism, which has led to some criticism. John Gruber at Daring Fireball noted that Anthropic posted about "how Claude marks AI-generated content" without fully explaining how the marking works. This is likely intentional — revealing the exact watermarking algorithm would make it easier to circumvent. But it creates a trust problem: how do we verify the watermark is working as described without understanding the mechanism? The Broader Watermarking Landscape Anthropic isn't alone in exploring text watermarking: Google: Has published research on SynthID-Text, a statistical watermarking approach OpenAI: Previously explored watermarking but reportedly faced internal debates about implementation Meta: Has discussed watermarking for Llama model outputs Academic research: Multiple approaches exist, from statistical frequency manipulation to cryptographic embedding The challenge all these approaches face is the tension between robustness (surviving editing, translation, paraphrasing) and stealth (not being obvious enough to avoid). Anthropic's approach seems to prioritize persistence — making the watermark survive the journey from Claude to wherever the text ends up. What's Next Anthropic's watermarking is being rolled out as part of compliance with EU regulations, which means it will likely become the standard for AI-generated text transparency in Europe. Other AI providers will need to implement similar measures to comply with the AI Act. For the broader AI ecosystem, this means: AI-generated text will become identifiable — at least when using compliant models like Claude Content authenticity verification will become a standard feature, not a premium add-on The cat-and-mouse game between watermarking and anti-watermarking will intensify Legitimate AI use cases may face friction from watermarking — a trade-off that regulators have accepted The era of invisible AI text watermarks is here. Whether they'll actually solve the problem of AI-generated content transparency — or just create a new set of technical and ethical challenges — remains to be seen.

You may have seen watermarks on AI-generated images to indicate that they were indeed made with AI. Now, even AI-generated text is being watermarked to make it easier for consumers to spot online content and comments that were written by AI. Anthropic, the company behind Claude, has signed the EU’s Code of Practice on Transparency of AI-Generated Content and explains how text watermarking works. Text watermarks are machine-readable but should be transparent to humans. “When a supported Claude model generates text, it weaves an imperceptible watermark directly into the text itself. You...

Anthropic on Tuesday announced that it is adding invisible watermarks to text generated by its artificial intelligence (AI) models. As per the company, the content will have a hidden signal that will be carried even when it is copied and pasted elsewhere. The move builds upon Anthropic’s decision to sign the European Union’s AI Act Article 50(2) Code of Practice on Transparency of AI-Generated Content. Anthropic says Claude models launched on or after August 2, 2026, will support the marking system from day one, and the feature is rolling out globally.

Anthropic joins a growing list of companies letting customers know if they’re consuming AI content.

Watermarks are invisible signatures that prove where a piece of content originated from.

Anthropic’s new watermarks will change how businesses worldwide use Claude—and how transparent they are about it.

Anthropic said in a support document Tuesday that its future AI models will put an invisible watermark in all text, identifying it as AI-generated. The change comes in response to new AI transparency laws in the EU, but Anthropic says it will apply in all countries where Claude models are used. No U.S. law currently requires AI-generated content to carry transparency disclosures or watermarks, although a growing number of states, California among them, have enacted AI transparency laws with labeling requirements. In the support document, Anthropic says it has signed the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content, which includes watermarking provisions, and explains how it plans to put those rules into practice. Claude models released after August 2, 2026, will generate watermarks within text at launch, Anthropic explained. Under EU law, AI providers will have until December 2, 2026 (about four months), to upgrade earlier or existing models to support the machine-readable marking requirement. Anthropic isn’t giving away too much information about how it will add the watermarks to the text, but it says the watermark will “travel with the text when it’s copied and pasted elsewhere.” More details will be included in “forthcoming technical documentation,” the company says. The models will also inject standard provenance data into generated images, video, and audio. That part of the plan will satisfy the requirements of California’s AI transparency law. Anthropic will use the Coalition for Content Provenance and Authenticity (C2PA) standard. According to the European Commission’s signatory list, about 190 companies and organizations had signed the Code of Practice on Transparency by the end of July 2026. Other than Anthropic, signatories include OpenAI, Google, Microsoft, Amazon, Mistral, Cohere, and others. Conspicuously absent are Elon Musk’s xAI and Mark Zuckerberg’s Meta.

Anthropic has begun labeling its models’ content as AI-generated to comply with new European Union regulations. The artificial intelligence (AI) startup noted the change on the support page for its Claude model Tuesday (Aug. 11), saying it came in response to the European Union’s AI Act’s Code of Practice on Transparency of AI-Generated Content. “As […] The post Anthropic Adds Watermarks to Denote AI-Generated Content  appeared first on PYMNTS.com.

New models will embed an invisible, machine-readable signal in AI-generated text, part of an industry-wide push to distinguish human writing from AI output.

Anthropic is introducing a Claude AI watermark that embeds machine-readable marks into text generated by supported Claude models. As per Anthropic, the supported Claude models will start embedding an invisible watermark in the text that they generate. The machine-readable mark is not supposed to be noticeable by the user and does not affect the semantics, […]

Anthropic announced Monday that new versions of its Claude AI models will embed an "imperceptible watermark" into the text they generate, a move that could make it harder to pass off AI slop as being created by humans. The post AI Slop Identified: Anthropic Adds Invisible Watermark to Claude’s Text Output appeared first on Breitbart.

The AI company signed the E.U. AI Act's transparency code and says new Claude models will mark content from day one of release

In compliance with EU rules, Anthropic's Claude will start watermarking AI-generated content.

Anthropic will invisibly watermark all of Claude's text and file output from 2 August to meet EU AI Act transparency rules, with the marking applied globally, whether or not the user is anywhere near Brussels.

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 metadata where supported," Anthropic says on a new Claude support page. The changes are invisible to human […]

Anthropic says Claude can now embed invisible watermarks in AI-generated content.

Anthropic’s new watermarking system quietly marks AI generated text and files, creating a new way to identify their origin across platformsContinue reading on Medium »