changes apply to Claude models<\/a> launched on or after August 2, 2026 and are being introduced across Anthropic’s products and services. The company says the system is part of its commitments under the European Union’s AI Act Article 50(2) Code of Practice on Transparency of AI-Generated Content.<\/p>\nThat means the change is less about putting an obvious \u201cmade by AI\u201d label on a response and more about creating a hidden trail that machines can potentially recognise.<\/p>\n
Claude’s watermark: How does it work?<\/p>\n
Unlike a visible label attached to a Claude response, the watermark is built into the generated text itself. Anthropic says readers should not notice any difference in meaning, quality or readability.<\/p>\n
There is a practical reason for embedding the mark in the text rather than tying it to the Claude interface. If a user copies a response and pastes it somewhere else, the signal can travel with the content. Anthropic says it may also survive certain forms of editing.<\/p>\n
![\"tech\"](\"https:\/\/images.firstpost.com\/dlxczavtqcctuei\/news18\/static\/images\/fp\/revamp_v1\/tech.svg\")
More from Tech<\/p>\n
The watermark is applied at the model level, so it is not dependent on whether someone accesses Claude through a particular app or service. Anthropic says supported models will retain the marking when used through platforms including AWS, Google Cloud and Microsoft Foundry.<\/p>\n
Files will use a separate system. For supported formats such as SVG, PNG and JPG, Claude will attach digitally signed provenance metadata using the C2PA standard, developed to establish the history and authenticity of digital content. This metadata can indicate that a file was processed by Claude and help identify tampering.<\/p>\n
However, the two systems should not be confused with an AI detector.<\/p>\n
A Claude watermark does not prove AI authorship<\/p>\n
Anthropic is explicitly warning against treating its watermark as a definitive answer to whether something was written by AI.<\/p>\n
Consider someone who gives Claude their own article and asks it to translate, summarise or proofread the material. The resulting text could contain a Claude watermark even though the underlying ideas and original writing came from the user.<\/p>\n
The reverse is also true. If a Claude watermark cannot be found, that does not establish that humans wrote the content. Anthropic says signals can disappear or become difficult to detect after extensive editing, paraphrasing or translation. Content combined with other writing may also lose the reliability of the signal, while very short passages may not contain enough material for detection.<\/p>\n
That makes the technology closer to a provenance tool than a digital lie detector. It can potentially provide evidence that content passed through Claude, but it cannot establish the origin of every idea or sentence.<\/p>\n
Anthropic is developing tools intended to let users and third parties check for its embedded watermarks and provenance metadata.<\/p>\n
The rollout will not cover every Claude model immediately.<\/p>\n
The broader shift points to a future in which AI-generated content may carry information beneath the surface \u2014 not necessarily visible to people reading it, but available to systems trying to establish where digital content came from.<\/p>\n","protected":false},"excerpt":{"rendered":"Anthropic is adding machine-readable markers to content generated by newer Claude models, giving text and files a hidden…\n","protected":false},"author":2,"featured_media":134576,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[32421,37869,67197,53,67406,67408,19043,66800,67407],"class_list":["post-135866","post","type-post","status-publish","format-standard","has-post-thumbnail","category-anthropic","tag-ai-act","tag-ai-detection","tag-ai-generated-text","tag-anthropic","tag-anthropic-claude-watermark","tag-c2pa-standard","tag-claude-models","tag-content-authenticity","tag-digital-watermark"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/135866","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=135866"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/135866\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/134576"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=135866"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=135866"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=135866"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}