Anthropic has recently clarified that the company will now be making text generated by supported Claude AI models. The AI startup recently updated its support page, where they clarified the adding of machine-readable marks to content generated by Claude. The marks will remain hidden from readers and are designed not to change the text’s meaning, appearance, or readability. Reports also indicate that the watermark could survive copying and pasting, as well as some forms of editing, allowing the origin of AI-generated content to be traced more easily. Anthropic clarified that the new rules are in compliance with the transparency rules under the European Union’s AI Act. The company is also reportedly developing tools that users, publishers, schools and other parties can use to check for these marks.
According to the Claude support page, the models will place a machine-readable watermark directly into the generated text. The watermark is designed to be invisible to people but detectable by software. Moreover, it’s also clarified that it will travel with the text when it is copied into another document.
Anthropic says that the watermark will also remain after some editing. However, the support page also clarifies that the watermark is not permanent in every situation, as heavy editing, paraphrasing, translation, or mixing Claude’s writing with other text can make the mark difficult or impossible to detect.
Claude can also process files such as PNG, JPG and SVG. For these files, Anthropic uses signed information about the file’s origin, following the C2PA standard. This information can indicate that Claude processed a file and can help show whether the file was changed later.
The marking applies worldwide to supported Claude models, which include Claude, Claude Code, Claude Cowork and Claude Tag, as well as Claude accessed through cloud services such as AWS, Google Cloud and Microsoft Foundry. New Claude models will support marking from launch.
Anthropic is also reportedly working to add marking to the text generated using the older models of the AI tool which were released before August 2, 2026.
Anthropic isn’t the first company to do so, as OpenAI and Google also previously did the same when they allowed the users to identify images generated using their models.
Anthropic says it is working on detection tools that can be used by organisations to check whether a text or a file contains a supported Claude mark. However, the details regarding the tool still remain under wraps.
The Claude support page also clarified that if a Claude mark is identified, then it should not be treated as proof that Claude wrote the entire piece, as many people use the tool for proofreading, translating, summarising or converting content. Hence, the content may also have been changed or combined with human writing afterward.
Also read: Google Pixel 11 series to launch this week: Date, India timing and what to expect
Moreover, the AI company also noted that the opposite is also possible. If you find a mark, then that does not prove that AI was not used to generate content, as older Claude models may not have marking support yet, and heavy rewriting or other changes can also remove or hide the watermark.