AI text watermarking visual guide
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 files starting Aug 2, 2026. The invisible marks travel with copied text, aiding AI-content detection.
Background
Anthropic announced it will embed invisible watermarks in text and files generated by its Claude AI models, complying with the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content. The watermarking applies globally to all new Claude models launched from August 2, 2026, and will be extended to existing models. This move is part of a broader industry trend where AI companies like Google and Suno are adopting watermarking to increase transparency and address concerns about AI-generated content being mistaken for human work.
How it unfolded
- Aug 2, 2026EU AI Act's Transparency Code takes effect, requiring AI companies to mark AI-generated content in machine-readable ways.
- Aug 11, 2026Anthropic announces it will watermark Claude-generated text and files, with new models supporting watermarking from launch.
- recentlyAnthropic confirms watermarking applies globally and will travel with copied text, potentially surviving some editing.
Who’s saying what
- Official
- Anthropic states the watermark is part of the text and will travel with it when copied, aiding transparency.
- Analysts
- Forbes notes the move could be a 'grand revelation' or a 'thunderous dud' in discerning AI content at scale.
- Caution
- Android Authority highlights uncertainty about how much editing might remove the watermark, raising questions about its robustness.
Still unverified
The exact technical implementation of the text watermark is not detailed by Anthropic, and its resilience to heavy editing remains unconfirmed.