ANTHROPIC
Copy, paste… watermark stays
Anthropic is adding watermarks to text generated by Claude to comply with new EU AI rules.
Under the EU AI Act’s Transparency Code, AI companies must mark AI-generated or edited content so other systems can identify it.
Anthropic says models released after 2 August will automatically watermark text and files, with older models set to follow. Files will use the C2PA standard.
The text watermark will stay with the content when it’s copied and pasted, and may remain after some editing. Anthropic hasn’t said how much editing would remove it.
In brief:
Claude-generated text will carry a detectable watermark.
The watermark can remain after copying, pasting or editing.
More AI companies are adding similar labels as regulation increases.
Everyone’s getting name tags
The feature will apply across Claude, Claude Code, Claude Cowork, Claude Tag and the Claude API.
Other companies, including Google, Meta, Microsoft and OpenAI, have also agreed to follow the EU rules.
The moment we’ve all been dreading. - MV