Anthropic’s November 3, 2026 platform release notes expanded text-watermark coverage to Claude Mythos 5, Claude Opus 4.7 and Claude Sonnet 4.6 on the platforms described in its marking documentation. The update extends an existing provenance effort to additional earlier models.
For organizations publishing or storing model output, the practical question is which outputs carry a mark and what that mark establishes. Coverage depends on the model and platform, and Anthropic documents additional conditions for particular generation settings.
Model and platform details affect coverage
Anthropic’s marking guide lists embedded text watermarks separately from Content Credentials attached to supported generated files. The two mechanisms provide different types of provenance evidence. A content review should identify which one applies to the asset being examined.
The current guide includes cloud rollout timing and a caveat for Sonnet 4.6 when certain sampling settings use non-default values. Teams should consult that live coverage table when recording the behavior of their own integration.
Detection access is a separate capability
Anthropic describes watermark detection as a private preview for eligible organizations, including groups with transparency-related responsibilities and enterprises reviewing their own outputs. The existence of a model-side mark therefore does not establish that every publisher can immediately access a general detection API.
Our operational interpretation is that a publisher should document the evidence available today. That can include the generating product, model version, original asset and editing history. A future detection integration can add another signal once the organization has an available, supported route to use it.
Provenance needs context
A marked output may have been translated, summarized or otherwise processed from material written by a person. A detection result alone cannot identify every contributor to the final work. Likewise, a missing detectable mark may reflect the asset’s processing history or the supported coverage.
For a publishing workflow, keep the original generated asset and the approved edited version linked in the internal record. Record material transformations such as translation or format conversion. This allows reviewers to explain how the final item was produced using evidence that accompanies the asset.
How to update a content workflow
Start by identifying the Claude models and products used in the organization’s existing content work. Map those against the published coverage table and record where a platform rollout is still in progress. Then decide who owns disclosure and provenance checks at publication time.
The November expansion makes provenance a relevant operational topic across a broader set of Claude integrations. A useful implementation connects current model coverage with asset history, available detection tools and the publisher’s own editorial practices, so a reviewer can explain the origin of a specific piece of content.