--- title: Claude watermarks your code now | daily.dev description: Anthropic updated its documentation to comply with the EU AI Act's Article 50, committing to embed machine-readable watermarks in text and signed provenance... canonical: https://daily.dev/posts/claude-watermarks-your-code-now-7pzs9dhfd twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Claude watermarks your code now | daily.dev og:description: Anthropic updated its documentation to comply with the EU AI Act's Article 50, committing to embed machine-readable watermarks in text and signed provenance... og:url: https://daily.dev/posts/claude-watermarks-your-code-now-7pzs9dhfd og:image: https://api.daily.dev/og/posts/7pzS9Dhfd.png og:image:alt: Claude watermarks your code now og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Claude watermarks your code now **[Theo - t3․gg](https://daily.dev/sources/t3dotgg)** · 31 min read · 1 upvotes · 0 comments ## Summary Anthropic updated its documentation to comply with the EU AI Act's Article 50, committing to embed machine-readable watermarks in text and signed provenance metadata in files generated by Claude models, including Claude Code, starting with EU-launched models on or after August 2, 2026. The piece explains why image watermarking (easy to embed, easy to destroy via compression) differs from text watermarking (harder to embed since text is already compressed, but the marks can persist through editing). It covers techniques like SynthID-style token sampling and Unicode homoglyph tricks, and argues all such watermarks are trivially removable via paraphrasing or Unicode normalization, meaning the policy will only catch the lowest-effort spammers rather than serious bad actors. The author argues human-content verification (via C2PA-style signing) is a more durable long-term solution than trying to flag AI content. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### When will Claude start watermarking AI-generated text and code under the EU AI Act? Claude models launched in the EU on or after August 2, 2026 will support machine-readable watermarking at launch. Generated text will carry an embedded, imperceptible watermark woven in at the model level, and generated files (like SVGs, PNGs, JPEGs) will include signed C2PA provenance metadata where supported. This applies across the API, Claude.ai, Claude Code, and other Claude surfaces, and Anthropic plans to extend marking to older models too. _Developers tracking how AI provenance rules affect their Claude Code workflow can follow the details on daily.dev._ ### Can AI text watermarks like Claude's or Google's SynthID be removed or bypassed? Yes, text watermarks are trivially removable. Paraphrasing the output with any unwatermarked LLM strips SynthID-style statistical watermarks because they depend on subtle vocabulary/token-sampling choices; replacing Unicode homoglyph characters with normal ones removes homoglyph-based marks; and public detection tools let anyone iterate until a passage no longer triggers a positive. Repos already exist offering scripts and agent skills that strip multi-vendor AI provenance marks from text and files. _Anyone evaluating AI content-provenance tooling can weigh these limitations by browsing related coverage on daily.dev._ ### Why is watermarking AI-generated text harder than watermarking AI-generated images? Images contain vast amounts of perceptually irrelevant data (millions of pixels), so subtle changes like shifting a color value every few pixels go unnoticed by humans but get destroyed easily by compression like converting a PNG to JPEG. Text is already highly compressed with no redundant bits to hide data in, so any watermarking must rely on statistical token-selection patterns (like SynthID) or Unicode tricks, both of which are cheap to detect and cheap to strip via paraphrasing or character replacement. _Engineers weighing how much to trust AI content-detection claims can dig into the technical tradeoffs via daily.dev._ ## Similar posts on daily.dev - [Text AI watermarks will always be trivial to remove](https://daily.dev/posts/text-ai-watermarks-will-always-be-trivial-to-remove-borbz7x9u) · sean goedecke · 6 upvotes · 4 comments --- Tags: [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic), [#ai-governance](https://daily.dev/tags/ai-governance) [View this post on daily.dev](https://daily.dev/posts/claude-watermarks-your-code-now-7pzs9dhfd) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Claude watermarks your code now","url":"https://daily.dev/posts/claude-watermarks-your-code-now-7pzs9dhfd","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/claude-watermarks-your-code-now-7pzs9dhfd"},"datePublished":"2026-08-14T10:15:48.998Z","dateModified":"2026-09-14T06:18:22.833Z","description":"Anthropic updated its documentation to comply with the EU AI Act's Article 50, committing to embed machine-readable watermarks in text and signed provenance...","image":"https://i.ytimg.com/vi/Be-NqsW-wuk/sddefault.jpg","thumbnailUrl":"https://i.ytimg.com/vi/Be-NqsW-wuk/sddefault.jpg","isAccessibleForFree":true,"articleSection":"Theo - t3․gg","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Theo - t3․gg","logo":"https://media.daily.dev/image/upload/s--UmX7IyU3--/f_auto/v1704628081/logos/t3dotgg.jpg","url":"https://daily.dev/sources/t3dotgg"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/claude-watermarks-your-code-now-7pzs9dhfd","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"claude,anthropic,ai-governance","timeRequired":"PT31M","video":{"@type":"VideoObject","name":"Claude watermarks your code now","description":"Anthropic updated its documentation to comply with the EU AI Act's Article 50, committing to embed machine-readable watermarks in text and signed provenance...","thumbnailUrl":"https://i.ytimg.com/vi/Be-NqsW-wuk/sddefault.jpg","uploadDate":"2026-08-14T10:15:48.998Z","duration":"PT31M","url":"https://api.daily.dev/r/7pzS9Dhfd","embedUrl":"https://www.youtube.com/embed/Be-NqsW-wuk"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Theo - t3․gg","item":"https://daily.dev/sources/t3dotgg"},{"@type":"ListItem","position":3,"name":"Claude watermarks your code now"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/claude-watermarks-your-code-now-7pzs9dhfd#faq","mainEntity":[{"@type":"Question","name":"When will Claude start watermarking AI-generated text and code under the EU AI Act?","acceptedAnswer":{"@type":"Answer","text":"Claude models launched in the EU on or after August 2, 2026 will support machine-readable watermarking at launch. Generated text will carry an embedded, imperceptible watermark woven in at the model level, and generated files (like SVGs, PNGs, JPEGs) will include signed C2PA provenance metadata where supported. This applies across the API, Claude.ai, Claude Code, and other Claude surfaces, and Anthropic plans to extend marking to older models too. Developers tracking how AI provenance rules affect their Claude Code workflow can follow the details on daily.dev."}},{"@type":"Question","name":"Can AI text watermarks like Claude's or Google's SynthID be removed or bypassed?","acceptedAnswer":{"@type":"Answer","text":"Yes, text watermarks are trivially removable. Paraphrasing the output with any unwatermarked LLM strips SynthID-style statistical watermarks because they depend on subtle vocabulary/token-sampling choices; replacing Unicode homoglyph characters with normal ones removes homoglyph-based marks; and public detection tools let anyone iterate until a passage no longer triggers a positive. Repos already exist offering scripts and agent skills that strip multi-vendor AI provenance marks from text and files. Anyone evaluating AI content-provenance tooling can weigh these limitations by browsing related coverage on daily.dev."}},{"@type":"Question","name":"Why is watermarking AI-generated text harder than watermarking AI-generated images?","acceptedAnswer":{"@type":"Answer","text":"Images contain vast amounts of perceptually irrelevant data (millions of pixels), so subtle changes like shifting a color value every few pixels go unnoticed by humans but get destroyed easily by compression like converting a PNG to JPEG. Text is already highly compressed with no redundant bits to hide data in, so any watermarking must rely on statistical token-selection patterns (like SynthID) or Unicode tricks, both of which are cheap to detect and cheap to strip via paraphrasing or character replacement. Engineers weighing how much to trust AI content-detection claims can dig into the technical tradeoffs via daily.dev."}}]} ```