Anthropic's Claude watermark vanishes if you touch one setting

Anthropic is extending its invisible text watermark to four more Claude models — Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6 — ahead of the EU AI Act's December transparency deadline, after every model released since August 2 shipped with the mark from day one.

On Opus 4.6 and Sonnet 4.6, the mark only applies when developers leave temperature, top_p and top_k at their defaults; a minor tweak to any one of them produces unmarked output, and Anthropic's own guidance to API customers who rely on the watermark for compliance is simply to keep those settings unchanged.

The gap matters because Opus 4.6 remains Anthropic's best writing model, and the watermark — a statistical pattern in word choice, built on a variant of DeepMind's SynthID text method — is one of the few tools meant to prove whether text came from Claude at all.