Anthropic is adding a hidden watermark to text produced by Claude and its other AI models, the company announced yesterday, to meet European transparency requirements.
The technology will work across Anthropic’s products. The company also said it will use the C2PA standard to identify AI-generated files.
According to Anthropic, newer models released after August 2 will automatically include the watermark (when the EU AI Act’s Transparency Code takes effect), while support for older models will follow. The policy applies globally to all content Claude generates or edits.
The watermark is built into the generated text rather than added as a visible label. That means it can remain attached when someone copies and pastes the content elsewhere. The tech startup also said the mark could survive some edits, although it has not explained how much a user would need to change the text to remove it.
Anthropic’s move reflects a broader shift across the AI industry. Companies are facing increasing pressure to flag AI-generated content as concerns over computer-written material roil the publishing industry.
AI music platform Suno recently announced similar plans for its songs, while Substack has partnered with Pangram to flag AI-generated writing.
Google, Meta, Microsoft and OpenAI have also backed the EU’s approach to AI transparency.