Claude's Invisible Watermark Conflicts with EU AI Act Labeling Exemptions

max_paperclips · x · 2026-08-12

Anthropic's recent move to add invisible watermarks to Claude's generated text conflicts with Article 50 of the EU AI Act. The regulation includes an exemption stating that anyone who reviews AI-assisted text and holds editorial responsibility for it does not have to label it as AI-generated.

However, the technical implementation of Claude's watermarking undercuts this exemption. For instance, if a user dictates a sentence in German and uses AI to translate it word-for-word into English, the text gets marked as AI-generated anyway, despite the thought being entirely the user's and the user holding editorial responsibility. This contradiction—where the law grants an exemption but the technology marks it regardless—raises concerns among users about compliance boundaries.

Related event: Claude's Watermarking Sparks User Backlash Over EU Act Conflict(2 posts)→

Original post →

More from Models

Models channel →