Anthropic adds text watermarks to four older Claude models, but on Opus 4.6 and Sonnet 4.6 a sampling tweak removes them
The Decoder reported on 31 October 2026 that Anthropic told API customers by email it will watermark text from Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6 to meet the EU AI Act’s December deadline, with no option to turn it off. On Opus 4.6 and Sonnet 4.6, however, the watermark applies only when the temperature, top_p and top_k sampling parameters stay at their defaults; changing any one of them avoids it.
According to the email as reported by The Decoder, Anthropic will watermark Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6 and Claude Sonnet 4.6; newer models such as Claude Opus 5.5 already carry the mark, and the rollout on Amazon Bedrock may take a few extra days. Anthropic describes the watermark as an “imperceptible pattern in word choice” that does not change meaning, quality or readability, adds no characters or tokens, does not affect pricing or request and response formats, barely slows output, and “contains no information about the user, their organization, or their conversations with Claude.”
The caveat concerns the two older models. Anthropic says Opus 4.6 and Sonnet 4.6 are watermarked only when developers leave all three sampling parameters, temperature, top_p and top_k, at their default values, so a small change to any one of them avoids the mark. The email warns: “If you rely on the watermark for your own compliance, keep these settings at their defaults.” The Decoder notes the email does not mention this caveat for Mythos 5 or Opus 4.7 and infers that the mark applies there regardless of sampling settings, which remains an inference.
A Claude Help Center article updated this week says Anthropic has signed the EU AI Act Article 50(2) Code of Practice on Transparency of AI-Generated Content and, in line with it, is adding watermarks to outputs from models released before 2 August 2026. Text watermarks are applied at the model level, so they are present whichever Claude product or surface produced the text and travel with it when copied; supported files such as PNG and JPEG get C2PA Content Credentials. The Decoder says the mark also works for these models on AWS, Google Cloud and Microsoft Foundry and uses a variant of Google DeepMind’s SynthID text method.
Watermark detection is in private preview, available on application to organizations eligible under EU law, such as regulators, law enforcement, media, fact-checkers, independent researchers and educational organizations, and to enterprises with their own obligation to verify watermarks; Content Credentials on files can be checked with the free Claude Content Checker. The Help Center also says developers who deploy Claude in their own products should independently assess what Article 50 requires of them.
Sources
- Anthropic’s watermark rollout leaves a loophole on two older Claude models — The Decoder
- How Claude marks AI-generated content — Claude Help Center (Anthropic)