Anthropic has published a detailed explanation of how its text watermarking works for Claude models.

The system embeds invisible statistical patterns in generated text to help identify AI-generated content, with plans to extend to earlier models.

AI-generated: summaries written by AI from the linked sources. How we use AI