Home / Technology / Anthropic Marks Claude AI Text with Invisible Watermarks
Anthropic Marks Claude AI Text with Invisible Watermarks
11 Aug
Summary
- Anthropic will embed invisible watermarks into AI-generated text.
- New EU AI Act mandates transparency for AI-generated content.
- Watermarks aim to make AI-generated content easily detectable.
Anthropic has announced a commitment to marking content generated by its Claude models with machine-readable data. This initiative is in response to the European Union's AI Act, which came into effect on August 2nd, 2026. The AI Act mandates increased transparency for artificial intelligence systems.
Newer Claude models will embed these watermarks from their launch, while existing models will receive support as a work in progress. For images, Anthropic will use the C2PA provenance metadata standard. Text watermarking involves weaving imperceptible marks directly into the generated content, aiming to preserve quality and readability while ensuring the marks travel with copied or edited text.
