Anthropic to Watermark AI Text, Including Claude, to Meet EU Rules

To adhere to the EU AI Act's Transparency Code, Anthropic will implement watermarking for text produced by its AI models, such as Claude, starting August 2. This regulation mandates that AI firms label AI-generated materials for easy identification. All models introduced post this date will inherently feature watermarking technology based on the C2PA open standard, which remains intact through certain edits and copies. The watermark will be integrated at the model level across all Claude offerings, including the API and Claude Tag, with support extended to previous models. The difficulty of removing the watermark through editing is still uncertain. Other firms, like Suno, Substack, Google, and Microsoft, are also complying with the EU guidelines.