Anthropic to Implement Invisible Watermarks in AI-Generated Text for Identification
Anthropic plans to introduce an imperceptible watermark in text generated by its AI model, Claude, starting from August 2. This measure aims to comply with the EU AI Act's transparency requirements and is part of a broader industry effort to address concerns over low-quality AI-generated content.
Anthropic will embed a machine-readable watermark in text produced by its Claude AI model, which will be undetectable to users but will accompany the text when copied or pasted.
The watermark is designed to comply with the EU AI Act's new transparency rules that took effect on August 2, requiring generative AI outputs to be identifiable.
This initiative comes amid growing concerns about the proliferation of low-quality AI-generated content across various platforms, prompting companies like Substack and YouTube to implement measures to differentiate human and AI-generated work.
While watermarking aims to help identify AI content, it may not fully prevent users from altering or removing the mark, and there are concerns about the implications of labeling all AI-generated content uniformly.
Anthropic is also working on extending this watermarking feature to older versions of Claude.