Anthropic announces new watermark system for Claude AI content

AI company Anthropic is implementing a new system to embed invisible watermarks within content generated by its Claude AI models. This move is a direct response to transparency regulations outlined in the European Union's AI Act.

The new feature will be applied across various Claude models and platforms, including Claude Code, Cowork, Claude Tag, and services like AWS, Google Cloud, and Microsoft Foundry. Anthropic stated that this change is intended to help identify when Claude has been involved in text generation, thereby complying with the EU's requirements.

Anthropic emphasized that the watermarking process does not alter the quality or content of Claude's outputs. Users will not be able to distinguish between watermarked and unwatermarked text. Furthermore, the watermark itself contains no personally identifiable information and does not reveal any details about specific individuals, organizations, or conversations.

Instead of embedding hidden code or characters, Anthropic explained that the watermark is created by subtly altering the statistical distribution of Claude's word choices. This results in a unique "fingerprint" for the generated text, serving as a statistical signature without impacting the readability or usability of the content.

“

Claude modelsinde uygulanan yeni filigran sistemi, Claude Code, Cowork, Claude Tag veya AWS, Google Cloud ya da Microsoft Foundry gibi sistemlerde de üretilen sonuçlarda da yer alıyor.