Anthropic watermark on older Claude models can be skipped by changing settings

Anthropic told API customers by email that it will add its text watermark, which it describes as an imperceptible pattern in word choice, to Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6, the-decoder reported Oct. 31. On Opus 4.6 and Sonnet 4.6 the watermark applies only when the temperature, top_p and top_k sampling settings are left at their defaults, so a change to any one of them avoids it; the email gives no such caveat for the other two models, per the-decoder. Anthropic cites the EU AI Act’s December deadline for the rollout and says the mark contains no information about users, their organizations or their conversations.