Anthropic has clarified how it will implement watermarking for text generated by its AI chatbot, Claude, in compliance with the EU AI Act. The watermark will create an undetectable pattern in responses, ensuring that AI-generated content can be identified without affecting the quality of the output. Users have expressed mixed reactions, with some canceling subscriptions over concerns about transparency and potential misuse.
The company plans to use the SynthID-Text method and will release a watermark detection API. While light editing may not fully remove the watermark, significant rewriting could alter the text's classification as AI-generated. Other AI developers are expected to adopt similar watermarking practices as part of the Code of Practice.
As watermarking becomes standard in AI, users should monitor how this affects content creation and sharing. Future developments may include enhanced detection tools and broader industry adoption, raising questions about transparency and user trust in AI-generated content.