Anthropic is watermarking text generated by Claude to comply with EU law (www.engadget.com)

Anthropic has announced its implementation of watermarking for text generated by its AI model, Claude, in compliance with the European Union's new AI transparency regulations. This watermarking method prevents any visible markers or hidden characters in the text; instead, it enforces a pattern that can only be decoded with a specific key. By utilizing a key method based on examples like the digits of pi, Claude will determine word choice systematically rather than relying solely on random selections, ensuring that generated text is traceable to the AI while maintaining output quality and efficiency. This development is significant for the AI/ML community as it sets a precedent for compliance with emerging legal frameworks regarding AI transparency and accountability. Anthropic's approach also acknowledges the limitations of watermarking, noting that it may not identify whether Claude merely edited content. Furthermore, the company plans to extend watermarking to images and utilize cryptographic notes for validation. This holistic approach to watermarking ensures that AI-generated content remains identifiable, which could help mitigate concerns over copyright and unauthorized use, all while adhering to evolving regulatory standards.