Claude models' text now carries a watermark

According to Anthropic's Claude Platform release notes, from 3 November 2026 text produced by Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 carries Anthropic's text watermark. The note does not explain how the watermark is read.

What does the note say?

The release note says two things. First, the text output of the three models above carries a watermark. Second, on Claude Sonnet 4.6, text generated with a non-default temperature, top_p or top_k might not carry it. So the mark is not present in every case.

What is a text watermark for?

Like the hidden marks in images, it is meant as a signal that helps work out afterwards whether a text was produced by an AI model. The release note does not describe how the mark is verified, how accurate it is or who can read it, so it cannot be concluded that a text can be reliably identified as AI-written.

For businesses that publish content

The practical takeaway: a business that drafts with AI should assume the origin of a text may be questioned later. That is not a reason to stop using AI; it is a reminder to build a routine in which you can explain at which stage AI was used and who approved the result, instead of trying to hide it. What matters is not avoiding the mark but verifying the text before publishing and taking responsibility for it. We summarised Google's view of AI content in a separate guide.