Anthropic unveils AI watermarks to comply with EU law

Anthropic is introducing an invisible watermark to text produced by its Claude chatbot, a move designed to comply with new European Union regulations on artificial intelligence. The company announced Monday that the feature will be applied to all Claude models launched on or after August 2, with older versions to be upgraded.

While the watermark will not be visible to humans and will persist through copy-pasting, Anthropic cautioned that it signifies only that text was processed by Claude, not that the AI was the original author. The company explained that Claude is often used for tasks like proofreading, translation, and summarization, meaning output can carry a Claude mark even if the underlying ideas originated elsewhere. Conversely, very short or heavily edited texts might not receive a watermark.

Anthropic stated that when Claude generates supported file types like .jpg, it will embed signed provenance metadata. This initiative aligns with the EU's call for AI-generated content to feature machine-readable markings, with potential fines for non-compliance. The watermarking will be implemented globally, not just within Europe.

This development comes as Anthropic joins approximately 190 organizations, including Google, Meta, Microsoft, Mistral, and OpenAI, in signing a voluntary EU code of good practice. Google has developed its own watermarking technology, SynthID, and is encouraging industry adoption of its standard. However, Anthropic has not yet released a detection tool for its watermarks, as recommended by the EU code.

“

Claude may not be the original author. People often use Claude to proofread, translate, summarise, or convert files. The output can carry a Claude mark even if the underlying ideas, text, or data originated from another source.