Claude announces invisible watermarks for text
Claude, the language model from Anthropic, has announced that it will embed invisible watermarks in any text it generates. The move is part of a broader effort by several AI labs to comply with the European Union’s AI Act, which requires transparency about AI‑generated content.
Key points
- Claude will embed invisible watermarks in all generated text
- Google’s SynthID marks images and videos at the pixel level
- EU AI Act drives watermark adoption across major AI labs
The article explains that invisible watermarks are the most robust form of marking, surviving edits, cropping, and screenshots. Google’s SynthID system, used by OpenAI and now by Claude, embeds a watermark at the pixel level for images and videos. For text, the watermark is hidden in the probability distribution of token choices, making it hard to detect without specialized tools.
While the EU mandate pushes companies like Anthropic, Meta, Microsoft, and Black Forest Labs to adopt watermarking, the article notes that xAI and Midjourney have not yet signed up. Detecting these watermarks requires different tools from each provider, and rewriting text can remove the watermark, limiting its effectiveness.
The story so far
3 episodes →- Claude announces invisible watermarks for textthis story
What is an AI watermark, and what does it actually prove?
Zapier Blog · 24 September 2026
Loading the full article…
This text was published by Zapier Blog and written by Harry Guinness. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Policy & Regulation
All →- OpenAI banned account after user discussed mass shooting plans · 5 src
- Accenture stock rises after $1 billion AI safety partnership with Anthropic · 8 src
- OpenAI agent accessed Australian Medicare portal in June, Albanese says · 43 src
- Meta’s AI agent could face trust issues over personal data, says Zuckerberg · 1 src
- UN security council to hear AI risk briefings from OpenAI, Anthropic and Hugging Face · 28 src
Comments
via GitHub Discussions