Anthropic Expands Text Watermarking to More Claude Models
November 3, 2026 · 2 min read
Also published in Norsk
Anthropic is extending its text watermarking to four additional Claude models. The company notified its API customers via email that Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 will receive the watermark starting November 3, 2026. The rollout on Amazon Bedrock may take a few additional days.
The watermark is described as an imperceptible pattern in word choice that does not alter the meaning, quality, or readability of responses. According to Anthropic, it does not add extra characters or tokens, does not affect pricing or request and response formats, and has a negligible impact on output speed. The company stated that the watermark contains no information regarding the user, their organization, or their conversations with Claude.
A loophole exists for users of the older Claude Opus 4.6 and Claude Sonnet 4.6 models. For these two versions, the watermark is only applied if the sampling parameters temperature, top_p, and top_k remain at their default values. Changing any of these values may result in outputs that do not carry the watermark. Anthropic advised customers who rely on the watermark for their own compliance to maintain the default settings. This restriction was not mentioned for Claude Mythos 5 and Claude Opus 4.7, suggesting the watermark functions there regardless of sampling settings.
Anthropic cited the December deadline of the EU AI Act as the reason for the watermarking. Models released after August 2, 2026, already include the watermark from the first day, while older models are being retrofitted. Because the marking is embedded at the model level, it applies across all platforms where the models are offered, including Amazon Web Services, Google Cloud, and Microsoft Foundry. The company uses a variant of the SynthID Text method from Google DeepMind, which modifies the random source during word selection to create a statistically detectable pattern. Customers cannot disable the marking.
To detect the watermark, Anthropic refers to a Detection API currently in private preview. The company also provides further details on its content labeling information page.