Skip to content
ClaudeDetect
Language: English

Does Claude watermark its text?

By Yang Zheng · Updated

Yes, Claude text is watermarked, but nobody outside Anthropic and approved organisations can check it, and this site does not read the watermark. ClaudeDetect works from writing style instead. Status as of 2026-11-10.

How the watermark works

It is Anthropic's version of the SynthID-Text approach that Google DeepMind published in Nature in 2024, run with Anthropic's own key. Nothing is added to the text: no hidden characters and no extra tokens. Whoever holds the key can ask how likely it is that a text was partly written by Claude. Anthropic applies it worldwide to comply with the EU AI Act, and signed the EU Code of Practice on Transparency of AI-Generated Content in July 2026.

Which models carry it

Images and other files get C2PA Content Credentials instead. See the Claude versions page for what each model is.

What it can and cannot prove

Who can verify it

Detection is a private-preview API for eligible organisations: regulators, law enforcement, media, fact-checkers, independent researchers, educational organisations, EU civil-society groups, and enterprises that must check their own Claude outputs under the Act. Anthropic plans to expand access and takes requests through its Claude Watermark Detector Access Request Form. Individuals cannot check text themselves, and Anthropic has not published detector error rates or a minimum text length.

Checking a text without the key

Anthropic names Pangram as detection software that uses a different method, because such companies do not have its key, and says checking for phrasing patterns is fundamentally different from checking for a watermark. That second kind of check is what ClaudeDetect does. It reports an AI verdict, the passages that carry the signal, and the writing-style similarity to Claude and other families, using Veread, Fileverity's text model. Accuracy figures are on the facts page.

Sources: Anthropic, How Claude's text watermarking works; Claude Help Center, How Claude marks AI-generated content.

Questions and answers

Does this site read Claude's watermark?

No. ClaudeDetect does not read or verify Anthropic's watermark. It estimates whether a text reads as AI-written and how close its style is to Claude.

Can I check a text for Claude's watermark myself?

Not today. Anthropic's watermark detection API is in private preview for eligible organisations such as regulators, media, fact-checkers, researchers and educational organisations.

Which Claude models watermark their text?

Models launched on or after 2 August 2026 carry the mark from launch. Anthropic's support article also lists earlier models in its own apps and API, including Opus 4.5 to 4.8, Sonnet 4.5 and 4.6, and Haiku 4.5.

Does editing remove the watermark?

A full rewrite removes it, and light editing probably does not. Short passages, factual passages and code carry less signal.

How can I check a text for Claude without the watermark key?

Use a style-based detector. ClaudeDetect reports how likely a text is AI-written and how closely it matches Claude's writing style.