Claude's new Scarlet Letter watermark is invisible — for now

Anthropic has announced its plans to watermark content processed by its models, a move that aligns with the European Union’s AI Act. This law mandates all AI system providers to watermark AI-generated or manipulated content, including audio, images, text, and video. The regulation applies to any AI model released after August 2, 2024, and allows providers a grace period until December 2026 to update previously released models.

Moving forward, all new models offered by Anthropic globally will mark AI-generated content "from day one," with text outputs carrying "invisible" embedded watermarks. Other generated files will include digitally signed provenance metadata, wherever supported. Anthropic is taking a proactive approach by applying watermarks to all processed content where possible, even though the EU does not require it for assistive functions like grammar correction or when it doesn’t substantially alter the user’s text or its meaning.

However, there's uncertainty about the thoroughness of this watermarking, as Anthropic hasn't released a detection tool yet. The company plans to share further details on how to detect these marks to comply with the EU's requirements.