Anthropic began rolling out AI text watermarking for Claude outputs to help comply with the EU Code of Practice
Deployment Provisional 72% confidence first seen
Coverage reports that Anthropic is rolling out text watermarking for its Claude models, using a method that encodes a watermark by steering token sampling with a secret, key-derived pseudo-random source. The articles also describe prior implementations by Google and mention public results suggesting no meaningful user-feedback difference in a large-scale test, along with the availability of a detector API to verify whether outputs contain the watermark.