Anthropic’s Claude to mark all AI content, including text
Anthropic is implementing machine-readable watermarks and provenance metadata in its Claude AI models to comply with the EU AI Act. These markings are designed to identify AI-generated text and files, though the company notes limitations regarding accuracy and potential tampering.
Why it matters
This represents a significant step in AI transparency and regulatory compliance, setting a precedent for how AI companies verify content origin.
As part of its requirements to comply with European Union regulations, Anthropic announced that its Claude models will carry identity markings in order to enable the labelling of AI content.
Anthropic has signed the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content, meaning that Claude models launched in the EU on or after August 2 will support machine-readable markings at the time of launch.
Older models, meanwhile, are subject to a transition period.
“Generated text will carry embedded watermarks, and generated files will include digitally signed provenance metadata where supported,” stated Anthropic in the blog post.
Of particular interest is how Anthropic will watermark AI-generated text content, which is usually easier to pass off as being human generated, than an AI photo or a piece of art. Anthropic stated this would be done through an “imperceptible watermark” that cannot be seen by the reader.
Get smarter about the news
Sign up free for a feed built around what you actually care about, Dive Deeper research on any story, and the full text of every article.
Create free accountAlready have an account? Sign in