From 2 August, new transparency rules under the EU AI Act require providers of generative AI systems to ensure that content produced by their models can be identified as AI-generated.
In practice, companies must embed markers of origin, such as digital watermarks or metadata, in text, images, video and audio. These markers may not be visible to users, but can allow other platforms to identify content created or modified using AI.
The next step involves platforms where the content is shared, such as social media. If they can detect the embedded watermark, they can display a label indicating that the content was generated or modified using AI.
Anthropic, for example, said it is adding an invisible watermark to Claude-generated text that remains detectable even after copy-pasting.
Major AI companies are already adapting their systems to the new requirements. Anthropic, for example, has announced an invisible watermark for Claude-generated text designed to remain detectable even after the text is copied and pasted elsewhere.
The EU rules aim not only to address misleading content, but also to make the origin of digital content more transparent, helping users identify when content has been generated by AI.
Source: Politico
