Key points
- Anthropic is adding its text watermark to Claude Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6; Amazon Bedrock may lag by a few days.
- On Opus 4.6 and Sonnet 4.6, the mark appears only if temperature, top_p and top_k stay at defaults.
- Anthropic cites the EU AI Act's December deadline; customers cannot turn the watermark off, and a detection API is in private preview.
- OpenAI's rival system, textGrain, is on by default only for ChatGPT and Codex users in the EU; API users can toggle it.
Why it matters for marketers: Teams generating copy through Claude's API should know that watermarking is now mostly unavoidable, that tuning sampling settings changes it on two models, and that quality and client concerns remain disputed.
Anthropic is extending its text watermark to four more Claude models: Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6. The company told API customers by email that newer models such as Opus 5.5 already carry the mark, and that the rollout on Amazon Bedrock could take a few extra days.
Anthropic describes the watermark as an imperceptible pattern in word choice that does not change meaning, quality or readability. It says the mark adds no tokens, leaves pricing and formats alone, and carries no information about users or their conversations. Because it is built in at the model level, it applies across cloud platforms hosting the models, and customers cannot disable it.
There is a catch on two models. On Opus 4.6 and Sonnet 4.6, the watermark applies only when temperature, top_p and top_k are left at defaults. Anthropic's own advice is that anyone relying on the mark for compliance should keep those settings unchanged. The email reportedly raises no such condition for Mythos 5 or Opus 4.7. This matters because Opus 4.6 is described as Anthropic's best writing model.
For marketers and content teams, the practical issue is workflow. If you tune sampling for tone or variety, check whether your output is still marked. Quality is also contested: blogger John Gruber argues that choosing words by a watermark key costs precision, and Artificial Lawyer flags risks for law firms whose clients ban AI use.
Watch for broader access to Anthropic's detection API, now in private preview, and for how other vendors handle defaults. OpenAI's textGrain, for instance, ships disabled for API customers.
Sources
This brief is our own summary. Read the full reporting at the source.
