Anthropic has implemented a subtle watermarking system for its Claude models, which embeds a detectable trace within the generated text itself. This method, based on Google DeepMind's SynthID-Text approach, influences word selection using a secret key rather than adding hidden characters or altering the text's meaning. The watermarking is applied to all text generated by Claude models released after August 2, 2023, and will be extended to services like Amazon Bedrock by November 10. While the detector is not publicly available, Anthropic states that significant edits, paraphrasing, or translations can diminish or remove the trace, and the mark only indicates that Claude processed the text, not that it authored it. AI
IMPACT This subtle watermarking could impact AI content detection and attribution efforts, particularly for regulatory compliance.
RANK_REASON This is a product feature update for an existing AI model, not a new model release or core research.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
