Claude to watermark AI-generated content
Archived — this story has rotated out of today’s deck. It is kept here in full.
The gist
Anthropic will watermark Claude-generated text and images starting Aug 2. The invisible marks aim to identify AI content but may fail after heavy editing.
Background
Anthropic announced it will embed invisible watermarks in text and images generated by its Claude AI models, starting with models launched on or after August 2. This move aligns with the EU AI Act's transparency code, which requires AI companies to mark AI-generated content. The watermark is designed to travel with the text when copied and pasted, helping to identify AI-generated content, but it may not survive heavy editing or paraphrasing.
How it unfolded
- Aug 2, 2026EU AI Act transparency code takes effect, requiring AI companies to mark AI-generated content.
- Aug 11, 2026Anthropic confirms watermarking on an updated support page, stating Claude models launched on or after Aug 2 will include watermarking.
- Aug 11, 2026Anthropic's announcement sparks online debate about credit for AI-generated work, with some users arguing humans should get credit.
Who’s saying what
- Official
- Anthropic says watermarking is to comply with EU regulations and will share detection details in future documentation.
- Public
- Some Reddit users argue that humans should get credit for AI-generated content, saying 'I gave the instructions, context, decisions, and countless refinements, Claude was the tool.'
- Caution
- Critics point out loopholes: heavy editing, paraphrasing, or translating can make the watermark undetectable, and its absence doesn't prove content wasn't AI-generated.
Still unverified
The effectiveness of the watermark against sophisticated editing or translation is not yet proven; Anthropic has not released detection tools or details on how to verify watermarks.