According to CoinMeta, analyst Jukan stated that Anthropic's decision to add invisible watermarks to the Claude model on the eve of IPO was a serious act of self-harm. Anthropic began to add machine-readable invisible watermarks to the text generated by support models for Claude, Claude Code, Cowork, API, as well as AWS, Google Cloud, and Microsoft Foundry. The new models will support this feature upon launch, and existing models will be gradually updated to include it as well. To comply with Article 50 of the EU AI legislation on transparency, Anthropic has chosen to embed the watermarks directly within the text content. These watermarks can be transmitted through copying and pasting and are applicable globally without affecting readability.