Does Anthropic really watermark Claude's text?

Yes, since 11 August 2026. Every Claude model launched on or after 2 August 2026 embeds an imperceptible mark in generated text, applied at the model level so it is present on every Claude surface, and applied worldwide rather than only in the EU. The driver is EU AI Act Article 50, which became enforceable on 2 August 2026. Anthropic states the mark travels through copy-paste and may persist through some editing.

Can the watermark actually be removed?

Anthropic's own documentation says the mark may not survive heavy editing, paraphrasing, translation, mixing with other content, or very short passages. Paraphrasing is on that list, and the rewrite here is a paraphrase pass. But Anthropic has not published its detection mechanism, so nobody outside Anthropic - including us - can currently verify whether a mark is present or gone. We will not claim a result we cannot demonstrate. When a detector is published we will test against it and publish what we find.

What does a detected mark actually prove?

Less than people assume. Anthropic states a mark indicates content may have been processed by Claude - not that Claude authored it. Text you wrote yourself and asked Claude to proofread carries the same mark as text Claude drafted from scratch. The mark cannot distinguish those two cases.

What are the two layers you clean?

Character-level and statistical. The character layer is zero-width codepoints, Unicode tag characters, variation selectors, exotic spacing and typographic punctuation - removed by exact codepoint matching, verifiable by re-scanning, no model involved. The statistical layer is the vocabulary and cadence patterns, plus whatever the watermark contributes; that can only be addressed by rewriting.

Can invisible characters really be hidden in text?

Yes, and they survive ordinary copy-paste. Zero-width characters, Unicode tag characters and variation selectors all render as nothing while remaining fully present in the data. Run the free scan on any text and you will see the exact count.

Does the rewrite change my meaning?

It is instructed not to. Facts, figures, names, dates, quotations, citations, URLs, paragraph structure and claim strength are all held fixed, and any sentence without a marker is returned verbatim. Read the result before you use it — that is true of any tool that touches your writing.

Will this defeat a specific AI detector?

We make no claim about any named detector, and no claim to verified watermark removal. Detectors are proprietary, they change without notice, and they disagree with each other. Anthropic's watermark has no public detector at all yet. What we state precisely: the character-level markers are removed exactly and verifiably, and the wording markers in your scan report are what the rewrite targets.

Why is the scan free?

Because it costs us almost nothing. It is exact character matching, not a model call, so there is no reason to charge for it. You only spend words when you ask for an actual rewrite.

What happens to my text?

It is processed to produce your result and saved to your history so you can retrieve it later. You can delete any entry. It is never used to train a model.

How is usage counted?

By word, against the input text. Scans are free and never counted. Each plan also caps the words in a single request, so long documents need a higher tier or splitting up.

Can I use this on text I wrote myself?

Yes, and plenty of people do. Human writing picks up invisible characters from PDFs, CMS editors and word processors, and the typographic punctuation cleanup is useful on its own. The scan will tell you whether there is anything there.

Does Claude's API output carry the watermark?

For developers: whether text from the Anthropic API is marked, which models, and what it means for products built on Claude.

Removing the watermark · 2 min read

Short answer: yes. Anthropic applies the mark at the model level, which means it does not matter which surface the text came from — the app, claude.ai, the API, or your own product built on top.

Why model-level matters

If watermarking were applied at the product layer, API output could plausibly be exempt. It is not. Anthropic's documentation states marking is applied at the model, so every route to that model inherits it.

For anyone shipping a product on Claude, that has direct consequences:

Text your app generates for end users carries the mark.
Your users cannot opt out, and neither can you.
The mark travels with the text when it is copied out of your product.

Which models

Every Claude model launched on or after 2 August 2026 marks generated text at launch. Support for models released before that date is described by Anthropic as "in progress".

So model choice currently affects this. If you are pinned to an older model, its output may not be marked yet — but "in progress" means that will change, and pinning to an old model to avoid marking is a plan with an expiry date.

What it means for your product

If you generate user-facing prose — marketing copy, summaries, drafts — that output is marked, and your users may not know. Whether you disclose that is a product decision, but the EU AI Act's transparency obligations sit with providers of generative systems, and depending on how you deploy, that may include you rather than only Anthropic.

If you generate structured output — JSON, code, short field values — the mark is far less relevant. Anthropic states very short passages lack enough text for reliable detection. A model returning a 12-token JSON object is not carrying a usable signal.

If you post-process output — summarising, templating, mixing with your own content — Anthropic lists "mixing with other content" among the things the mark may not survive. Heavy templating likely degrades it as a side effect.

What you cannot do

Check for it. Anthropic has not published a detection mechanism, so you cannot audit your own pipeline's output for the mark, and you cannot verify whether your post-processing removed it. That is an unusual position to be in as an engineer: a signal you know is present, that you cannot measure.

Practical notes

Do not build a feature that promises watermark-free output. You cannot test it, so you cannot support the claim.
Do strip the character layer. Claude output contains em dashes, curly quotes and U+2026 heavily. If your product writes to a system that expects ASCII, or you care about consistent typography, normalise it. That part is deterministic and testable.
Watch for the detector. When Anthropic publishes its detection documentation, anything you assumed about your pipeline becomes measurable. Design so that finding out does not break a promise you already made.

Sources: Anthropic Help Center

Does Claude's API output carry the watermark?

Why model-level matters

Which models

What it means for your product

What you cannot do

Practical notes

Related

How to check if text has Claude's watermark

Claude's watermark and Turnitin: what schools can actually see

How to remove Claude's watermark from text