![\"FRANCE-TECHNOLOGY-SCIENCE-IT-AI-ANTHROPIC\"](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/08\/1786606220_408_0x0.jpg\")
<\/p>\n<\/p>\n
This photograph shows a figurine in front of the logo of the AI assistant “Claude” built by the US artificial intelligence safety and research company Anthropic during a photo session in Paris on February 13, 2026. (Photo by JOEL SAGET \/ AFP via Getty Images)<\/p>\n
AFP via Getty Images<\/p>\n
If you\u2019ve been using Claude to draft any kind of writing, that text will now carry a watermark you can\u2019t see \u2014 at least just yet.<\/p>\n
Earlier this week, Anthropic published a support page <\/a>explaining that Claude models launched from August 2 onwards embed an invisible watermark onto every patch of text they create.<\/p>\n \u201cAs AI-generated content becomes commonplace, greater transparency and signals about where content comes from can give people useful context about the information they consume,\u201d the firm stated in the post. <\/p>\n Images, SVGs and other files will be treated somewhat differently.<\/p>\n \u201cWhen Claude generates a supported file type, such as a .svg, .png, or .jpg, it will attach signed provenance metadata,\u201d Anthropic stated. <\/p>\n This metadata would follow the Coalition for Content Provenance and Authenticity<\/a> open standard, the same system Google and Adobe use. If a signed metadata label is present, it signals that a file was processed by Claude and lets you detect whether the file has been tampered with.<\/p>\n The marks apply across all Claude surfaces people use \u2014 the API, Claude Code, Claude Cowork and deployments through AWS, Google Cloud and Microsoft Foundry. Older models are being retrofitted with no stated timeline just yet.<\/p>\n The EU Law<\/p>\n The catalyst for the move is the EU AI Act\u2019s Article 50<\/a>, which became enforceable on August 2, and requires generative AI providers to specifically mark outputs in machine-readable formats so that downstream users, regulators and platforms may be able to determine AI-generated content. Fines for non-compliance can reach up to \u20ac15 million or 3% of a company\u2019s total worldwide annual turnover, whichever is higher.<\/p>\n Anthropic signed the EU Act\u2019s Code of Practice on Transparency of AI-generated Content<\/a>, which is a voluntary compliance framework that confers a presumption of meeting the Article 50 standard. As of late July, almost 200 companies had signed, including Microsoft, Google, Meta and OpenAI. <\/p>\n Google has been watermarking its AI-generated images since 2023 and hasw extended that to text, audio and video. Elon musk xAI has not signed. Meanwhile, OpenAI has reportedly had the capacity to watermark ChatGPT text for years, but has not done so \u2014 reporting from the Wall Street Journal<\/a> attributed that to concerns including false positives and competitive risk. That could change after Anthropic\u2019s move. <\/p>\n How does it work?<\/p>\n Interestingly, Anthropic is applying the requirement globally, not just to its EU users.<\/p>\n The text watermark works by statistically biasing Claude\u2019s word choices according to a key held by Anthropic. Individual choices look unremarkable, but across enough text, the pattern becomes detectable. <\/p>\n \u201cBecause the watermark is part of the text, it will travel with the text when it\u2019s copied and pasted elsewhere, and may persist through some editing,\u201d the firm stated. \u201cWatermarking will be applied at the model level, which means it will be present no matter which Claude product or surface the text comes from.\u201d<\/p>\n For files, as noted earlier, the C2PA meta data works differently, wherein a digital signed record of provenance is attached to the file. That signature can be checked to verify whether the file has been tampered with since Claude generated it. It can be stripped through re-saving or format conversion, which is why the two systems are meant to complement each other.<\/p>\n Anthropic also said it will release detection tools that allow users and third parties to check whether a piece of text or a file carries a Claude watermark. <\/p>\n It may be worth nothing that a detected watermark means that content may have been assessed or processed by Claude, not that Claude wrote it. Someone who used Claude to proofread, translate or summarise their work may produce output that carries the mark, even though the underlying text is their own. The mark would say that Claude touched it, not that Claude made it. <\/p>\n At the same time, the absence of a mark is not definitive confirmation of human authorship. And there will be no detectable signal on content generated from older models, very short passages or heavily paraphrased text or files whose metadata was stripped through screenshots or format changes.<\/p>\n Not Without Controversy<\/p>\n