On November 16, Anthropic extended text watermarking to Claude Opus 4.6 (claude-opus-4-6), Opus 4.5 (claude-opus-4-5-20251101), Sonnet 4.5 (claude-sonnet-4-5-20250929), and Haiku 4.5 (claude-haiku-4-5-20251001). The change applies worldwide on supported Claude products and APIs, including partner platforms on AWS, Google Cloud, and Microsoft Foundry. Amazon Bedrock may take a few additional days to complete the rollout.
Anthropic says the mark is embedded in the generated text, so it travels with a copy and may survive some edits. For these four models, non-default values for temperature, top_p, or top_k can make the watermark unreliable. Detection remains in private preview for eligible organizations, so the rollout expands the amount of marked text while access to verification is still limited.
