Anthropic details Claude watermarking to comply with EU AI Act

Anthropic published a blog post on August 15, 2026 detailing how it will watermark text generated by its Claude chatbot. The move is required to comply with the EU AI Act’s Transparency Code, which requires AI companies to use systems that make AI-generated content identifiable.

The watermark is invisible to readers but detectable with a key. Watermarking does not affect output quality, Anthropic said.

Anthropic will use the SynthID-Text approach outlined by Google DeepMind in 2024 and plans to release a watermark detection API. The technique differs from AI detection tools from Pangram, which look for writing patterns.

Light editing is unlikely to remove the watermark completely, but a complete rewrite will. If Claude only lightly edits a human author’s text, nearly all the words remain human-written and there is little for the watermark to attach to.

Code will carry less of a watermark because the model must produce working code, though comments within code can be watermarked.

Other major model developers have signed the same Code of Practice and will implement their own watermarks, Anthropic said.