Anthropic has revealed a previously unpublished detail: starting from August 2, 2026, new Claude models launched in the European Union will support machine-readable watermarking of generated text from their release. This applies to future releases—no models have been released after this date yet, the latest currently remains Claude Opus 5, released on July 24. This is stated in a Help Center article updated on August 10, 2026. The rule will apply not only in the EU, but everywhere Claude is offered—in the API, the Claude app, Claude Code, Claude Cowork, Claude Tag, as well as with partners like AWS, Google Cloud, and Microsoft Foundry.
Text and files are two different mechanisms
For text, Anthropic will use invisible watermarks that will be woven directly into the model's response, without changing its meaning, quality, or readability. Since the marker will become part of the text itself, it will persist during copy-pasting and partially—even after editing. This mechanism is not related to the C2PA standard.
For files—for example, .svg, .png, and .jpg—a different, already operational mechanism applies: Claude attaches signed provenance metadata according to the open C2PA (Coalition for Content Provenance and Authenticity) standard. Such a signature indicates that the file was processed by Claude and allows detection of tampering with the metadata after the fact. Support for this feature depends on the specific platform.
Old models without text labels
The text rule only applies to new versions of Claude. Models released before August 2, 2026—meaning all currently available models, including Claude Opus 5—are in a transitional period and do not embed the text watermark. Anthropic says it is working on adding watermarking support for them as well, and documentation will be updated as it becomes available.
Text cannot be verified yet
Anthropic states that it is working on separate tools that will allow users and third parties to detect text watermarks. The company promises to reveal details of the detection mechanism in technical documentation later—such a tool does not currently exist. Websites like c2pa.org or contentcredentials.org are not suitable for this: they verify metadata in files, not text watermarks.
An important nuance: even when detection becomes available, the presence of a watermark will not be definitive proof that the author of the text is specifically Claude, as the model is often used for editing, translating, or summarizing others' content. The absence of a watermark will also prove nothing.
AI Opinion
From the perspective of macroeconomic regulation, the coincidence of dates does not seem accidental. The European Union, from August 2, 2026, has introduced the mandatory requirements of Article 50 of the AI Act on labeling AI-generated content, and the European Commission already on July 20 approved the final clarifications for this norm, providing for fines of up to €15 million or 3% of the company's global turnover for violations of transparency. The launch of Claude's text watermark precisely on this day, it appears, is not a technical coincidence but a compliance response to a specific regulatory deadline.
A technical aspect not revealed in the article is the resilience of such watermarks to paraphrase and translation via another model. Industry experiments with text markers have shown that paraphrasing significantly reduces detection accuracy, and translating text to another language often removes the marker entirely. Whether Claude's invisible watermark will remain detectable after processing by a third-party AI is a question to which only the company itself currently knows the answer.
end-content
