Anthropic Adds Invisible Watermark to Claude Text: AI Transparency and the Stigma of Machine Writing

Anthropic has introduced an imperceptible watermark within the text generated by its language model, Claude, allowing AI detection systems to recognize passages created by AI. This development, announced last week, is in accordance with the EU’s AI Act, which mandates identifiable AI-generated content. Unlike current methods, Claude’s watermark functions at the word level, making it difficult to alter text to hide AI authorship. Some users have expressed their discontent, comparing the watermark to a “digital tattoo” and pointing out the stigma surrounding AI usage. While some critics question its effectiveness and potential impact on writing quality, supporters argue it enhances transparency, reflecting Anthropic’s commitment to training on human-written content and ensuring output provenance.