In brief
Anthropic’s confirmed rollout is based on release timing, not a permanent public list of model names. Claude models launched in the EU on or after August 2, 2026 support marking at launch.
Anthropic says it is also adding marking to eligible models released before that date. Coverage can change, so this page records the reviewed rule and date rather than freezing a list that could become inaccurate.
The confirmed Claude watermark rollout
Anthropic says Claude models launched in the EU on or after August 2, 2026 support machine-readable marking from launch. That is the clearest current rule for newly launched covered models.
Models released before that date are part of a transition. Anthropic says it is working to add marking support to eligible earlier models rather than claiming every historical model already has identical coverage.
Why this page does not publish a permanent model list
A static list could become wrong as Anthropic adds support to earlier models or launches new ones. The official rollout statement is about timing and eligibility, while individual platform features can also vary.
This guide therefore does not infer coverage from a model family name or assume that an older snapshot remains current. Check the review date above and consult Anthropic’s source for changes after that date.
Where supported model marking applies
For models that support marking, Anthropic says the marks apply worldwide wherever Claude is offered. Its documentation includes Claude products, Claude Platform access, and supported access through listed cloud partners.
That worldwide statement does not mean every output type has the same marking method. Generated text uses embedded watermarking, while supported files can use signed provenance metadata, subject to platform and feature support.
Model support does not guarantee a detectable result
Even when a model supports marking, a particular passage may be too short or may be heavily edited, paraphrased, translated, or mixed after generation. Those are separate limitations from whether the model supports marking in the first place.
A future detector must account for both coverage and the condition of the submitted content. No detected signal would not establish that Claude was never used.
How to check for later updates
Use the visible “Last reviewed” date to judge whether this guide reflects the period you are researching. Anthropic’s official marking documentation is the primary source for later model launches, transition updates, and platform changes.
This page should be revised when confirmed rollout information changes. It will not add model names based on inference, marketing language, or unverified detector claims.
Claude Model Watermark Coverage: FAQ
Do all Claude models have text watermarks?
Anthropic describes marking for supported models. New covered models launched in the EU on or after August 2, 2026 support it at launch, while eligible earlier models are being updated.
Why are individual model names not listed here?
A fixed list can become outdated during the rollout. This page reports Anthropic’s confirmed timing rule and review date instead.
Does support apply only in the EU?
No. Anthropic says that when a model supports marking, the marks apply worldwide wherever Claude is offered, subject to platform and feature support.
Does a supported model guarantee every passage can be detected?
No. Very short or substantially changed text may not have a detectable signal, and public technical detection guidance is still forthcoming.