Anthropic will watermark all Claude-generated text and images globally

Anthropic announced on August 11, 2026 that it will embed invisible watermarks in all text generated by Claude models and apply C2PA metadata to image files. The decision is a response to the EU AI Act’s Transparency Code, which took effect on August 2, and the change applies globally — not only to European users.

The text watermark works by nudging the model’s word choices according to a hidden pattern while keeping the text readable. The watermark travels with the content when it is copied and pasted, and may survive some editing. For writers who use Claude-based tools to draft, edit, or refine text, the practical question is what this means for how their published work will be characterized.

Anthropic was explicit about an important limitation: a detected watermark shows only that Claude processed the text at some point, not who originated the ideas or how much was written by a human. A writer who uses Claude to fix a grammatical error or translate a paragraph produces text that may carry the mark even if the substantive work was entirely their own.

The change affects Claude’s API, the chat app, Claude Code, Claude Cowork, Claude Tag, and access through AWS, Google Cloud, and Microsoft Foundry. Older models will be updated over time. Other major AI providers — Google, Meta, Microsoft, OpenAI — have committed to equivalent transparency standards under the same EU framework.

For writers and editors, the most immediate implication is disclosure. Platforms that scan for AI involvement, such as Substack’s Pangram integration, may now surface Claude-processed content that the author does not think of as AI-generated. The watermark does not change what was written; it changes how detection tools will characterize it.