Anthropic to add invisible watermarks to AI-generated text to meet EU rules

US artificial intelligence firm Anthropic has announced it will begin embedding invisible watermarks into text generated by its chatbot, Claude. This move is intended to comply with upcoming European Union regulations concerning artificial intelligence and its outputs.

The watermarks, embedded as metadata, will be imperceptible to users but will remain attached to the text even if it is copied and pasted elsewhere. Anthropic clarified that the watermark does not definitively prove Claude created the text, but rather indicates it was either generated or edited by the AI chatbot.

According to the company, this feature is already integrated into Claude versions released after August 2, with older versions scheduled for updates. This initiative directly addresses EU requirements that mandate machine-readable labels for AI-generated content, including text, images, and videos. Non-compliance could result in significant fines.

Anthropic joins a growing list of tech companies, including Google, Meta, Microsoft, Mistral, and OpenAI, that have signed a voluntary code of conduct related to AI. However, the EU's labeling requirements specifically apply to AI used for professional purposes, not for private individuals. Meanwhile, Google reported that its own AI chatbot, Gemini, has surpassed one billion monthly users.

“

The watermark does not necessarily mean that Claude created the text. It is possible that it was only edited or proofread by the AI chatbot.