Anthropic will begin labeling AI-generated content created by Claude, Claude Code, and the company’s API, as well as Claude models via third-party services, according to The Register.
Text generated by future Claude models will be given an invisible watermark. According to Anthropic, the watermark is embedded directly into the generated text without affecting its meaning or readability. For files, Anthropic will instead use signed metadata in accordance with the C2PA standard.
The effort is part of the company’s plan to comply with the EU’s AI Regulation and transparency requirements regarding AI-generated material. However, the labeling will be implemented globally for all users.
At the same time, Anthropic noted that the technology has its limitations. A label does not guarantee that material was created by Claude. Similarly, the absence of a label is not necessarily proof that the material is not AI-generated.