Anthropic announced plans to add watermarks to text produced by its AI models.

  Anthropic has announced that its AI models, including Claude, will now include watermarks on generated text to adhere to European regulations. This update is detailed on their revised support page. The EU AI Act‘s Transparency Code, effective from August 2, mandates that AI-generated or edited content must be marked in a way that makes it identifiable by other systems. Anthropic confirmed that all models released after this date will automatically employ technology to watermark both text and files, utilizing the C2PA open standard for files. The company is also working on extending this feature to older models. The watermark will accompany the text when copied or pasted and might still be visible after some editing. As noted on the support page, watermarking is applied at the model level, ensuring its presence regardless of the Claude product or source. However, the specifics of how much editing is required to remove the watermark remain unclear. Anthropic has been contacted for clarification, and updates will follow upon response. The watermarking will affect various products like the Claude platform API, Claude Code, Claude Cowork, and Claude Tag. Many platforms are now adopting similar measures in response to user backlash and to meet regulatory requirements. Recently, AI music platform Suno announced plans to watermark its tracks following legal challenges. Last month, newsletter service Substack collaborated with Pangram to mark AI-generated content, a move highlighted by CEO Chris Best amidst concerns over AI misuse. In addition to Anthropic, companies like Black Forest Labs, Google, Meta, Microsoft, OpenAI, and Synthesia have also committed to complying with the EU’s regulations. has announced that its AI models, including Claude, will now include watermarks on generated text to adhere to European regulations. This update is detailed on their revised support page. The EU AI Act‘s Transparency Code, effective from August 2, mandates that AI-generated or edited content must be marked in a way that makes it identifiable by other systems. Anthropic confirmed that all models released after this date will automatically employ technology to watermark both text and files, utilizing the C2PA open standard for files.

   The company is also working on extending this feature to older models. The watermark will accompany the text when copied or pasted and might still be visible after some editing. As noted on the support page, watermarking is applied at the model level, ensuring its presence regardless of the Claude product or source. However, the specifics of how much editing is required to remove the watermark remain unclear. Anthropic has been contacted for clarification, and updates will follow upon response. The watermarking will affect various products like the Claude platform API, Claude Code, Claude Cowork, and Claude Tag. Many platforms are now adopting similar measures in response to user backlash and to meet regulatory requirements.

   Recently, AI music platform Suno announced plans to watermark its tracks following legal challenges. Last month, newsletter service Substack collaborated with Pangram to mark AI-generated content, a move highlighted by CEO Chris Best amidst concerns over AI misuse. In addition to Anthropic, companies like Black Forest Labs, Google, Meta, Microsoft, OpenAI, and Synthesia have also committed to complying with the EU’s regulations.