Can you detect or remove Claude's text watermark?
Not with a generic hidden-character scan. Anthropic describes Claude's new text watermark as an imperceptible, model-level signal and says third-party detection details are forthcoming. Until those details exist, a responsible tool must label Claude as not testable.
The text checker can find literal Unicode and exact C2PA text credentials, but it does not call either one a Claude watermark. Anthropic's promised detector is the missing authority.
What Anthropic has announced
Anthropic says new Claude models launched in the EU on or after August 2, 2026 support machine-readable marking at launch. The marking applies worldwide to supported models across Claude, Claude Platform, Claude Code, Claude Cowork, Claude Tag, cloud partners, and other supported surfaces. Existing models are in transition.
| Claude output | Announced marking | What is public |
|---|---|---|
| Generated text | An imperceptible watermark woven into text at the model level. | Product scope and limitations; detailed detector mechanics are forthcoming. |
| Supported files | Digitally signed provenance metadata using C2PA. | The open C2PA standard can be validated by compatible readers. |
Why this is not a zero-width-character claim
Anthropic says the watermark is applied at the model level, travels with copied text, and may persist through some editing. It does not identify zero-width spaces or another public Unicode payload as the mechanism. Therefore, finding U+200B—or finding nothing—does not establish whether the Claude watermark is present.
What Check AI Watermarks reports today
| Claude-related layer | Can check here? | Result language |
|---|---|---|
| Announced Claude text watermark | No | Not testable without Anthropic's detector. |
| Literal hidden characters | Yes | Exact code point and cleanup policy; not attributed to Claude. |
| C2PA 2.4 text wrapper | Yes, structurally | Exact wrapper decoded; not relabeled as Claude. |
| C2PA structured-text block | Yes, structurally | Manifest reference decoded locally; external URLs are not fetched. |
| C2PA in PNG or JPEG | Yes | Manifest, asset integrity, signature, and configured trust are checked. |
| Claude-generated SVG credential | Not in this release | SVG is outside the current upload formats. |
Can the Claude watermark be removed?
No Claude-specific removal method can be verified from the public information available today. Rewriting or translation may reduce detectability according to Anthropic's limitations, but that is not the same as a deterministic remover and it changes the text. This app will not claim success it cannot measure.
What a negative result means
Anthropic says a missing detectable mark does not prove that content was not generated or processed by Claude. Possible reasons include an older model, short output, heavy editing, paraphrasing, translation, mixing with other text, or an unsupported platform or feature. A zero in this app is narrower still: no supported literal signal was found.
Primary sources and related checks
- Anthropic: How Claude marks AI-generated content
- How C2PA image validation works
- What the local text checker can verify
Frequently asked questions
Does Claude add a watermark to generated text?
Anthropic says models launched in the EU on or after August 2, 2026 support machine-readable marking at launch, and that embedded watermarks apply to text from supported models wherever Claude is offered. Older models are being transitioned.
Can a hidden-character checker detect Claude's watermark?
Not based on the information Anthropic has published so far. Anthropic calls it a model-level watermark that may persist through some editing and says technical detector documentation is forthcoming. A Unicode scan should not relabel itself as Claude detection.
Can this tool remove the Claude text watermark?
No verified Claude-specific removal is claimed. The tool can separately remove supported literal Unicode and C2PA text credential forms, but Anthropic has not said that any of them are its model-level text watermark.
Does no Claude watermark mean the text is not from Claude?
No. Anthropic lists unsupported older models, heavy editing, paraphrasing, translation, mixed text, very short passages, and unsupported surfaces as reasons a Claude mark may not be detectable.