Claude AI content marking is Anthropic’s answer to the European Union’s new transparency regime, though it is not a single universal watermark. The company says Claude models launched in the EU on or after August 2, 2026 will carry embedded marks in generated text and attach signed provenance metadata to supported files from launch. The policy will apply worldwide for supported models, not only to European users.
The rollout matters to anyone publishing or reusing Claude output: a positive detection result may show that Claude handled content, while a missing result settles very little. Anthropic has signed the EU AI Act Article 50(2) Code of Practice on Transparency of AI-Generated Content, according to its support documentation.
How will Claude mark AI-generated text and files?
For text, Anthropic says it will embed an imperceptible watermark during generation. The company says the signal does not change a response’s meaning, quality or readability. Because it is embedded in the text, Anthropic says it will travel when users copy and paste material and may remain after some editing.
Files get a different treatment. When Claude generates supported file types, including SVG, PNG and JPG, it will attach digitally signed provenance metadata based on the C2PA open standard. This is metadata associated with the file, not a claim that Claude is putting a watermark into image pixels. Anthropic says a signed provenance label indicates that a file was processed by Claude and can reveal whether it has been tampered with.
That distinction is not cosmetic. Text and file signals fail in different ways. Anthropic warns that file metadata can disappear through format conversion, re-saving or screenshots. A text mark may no longer be detectable after heavy editing, paraphrasing, translation or mixing with other writing. Very short passages can also lack a reliable signal.
Which Claude products will carry the marks?
Anthropic says supported-model output will be marked across its Claude consumer product, Claude Platform API, Claude Code, Claude Cowork and Claude Tag. Text watermarks will also apply when supported models are reached through AWS, Google Cloud or Microsoft Foundry. Signed provenance metadata may not be available on every partner platform because the necessary file features differ.
The promise is prospective for new EU launches after August 2. Anthropic says it is still adding marking support to models released before that date. So users should not assume every Claude response or file already carries a detectable marker.
Can a Claude mark prove who created content?
No. Anthropic says detection would indicate content may have been processed by Claude, not that Claude originated every idea, source or underlying piece of data. A user might ask the system to proofread, translate, summarize or convert material created elsewhere, and the resulting output can still bear a Claude mark.
The inverse is equally important: Anthropic says no detected mark does not establish that something was not generated or processed by AI. It could have come from an older Claude model, an unsupported product or file type, or content whose detectable signal was weakened or stripped. Anthropic says it plans tools for users and third parties to check text marks and provenance metadata, but has not yet detailed the technical detection approach.
This story draws on original reporting from Tom's Hardware.
