The company is also working to update existing models during the transition period permitted by the new legislation.
These measures are intended to help users and regulators distinguish between human-generated content and artificial output, a key mandate of the EU's regulatory framework.
Anthropic confirmed that these tracking mechanisms will be applied globally across all its platforms, including the Claude API and third-party cloud computing services like AWS, Google Cloud, and Microsoft Foundry.
While the company claims these marks will not alter the meaning or quality of the text, the move reflects a broader industry shift toward infrastructure that prioritizes legal accountability even as it introduces potential friction for customers who prefer unidentifiable AI assistance.
Despite these efforts, the effectiveness of the technical mechanisms remains a point of debate.
Anthropic will use the C2PA standard—a technical protocol for certifying the source and history of digital media—for file metadata, yet existing open-source tools can already remove such signatures.
Furthermore, the company acknowledged that while the text watermarks are designed to persist through copying and pasting, they are not a definitive guarantee of content origin.
Anthropic expects to release further documentation on how these marks can be detected as part of its ongoing compliance with the EU's disclosure laws.