Anthropic embeds hidden 'watermarks' in Claude text for EU transparency
Technology
Anthropic is rolling out hidden "watermarks" in text created by its Claude AI models.
These watermarks won't change how the text looks or reads, but they'll make it possible to estimate whether Claude was involved in generating the text, helping meet European Union rules about AI transparency.
Claude users see no changes
You won't notice any difference in speed, creativity, or cost when using Claude.
Even older versions (from before August 2, 2026) will get this update soon, so eventually text from future Claude models will have these invisible watermarks built in.