NewOctober 5, 2026: OpenAI announces text watermarking for ChatGPT models. Learn more →
Clean AI WatermarkClaude models launched since August 2026 leave a SynthID-based watermark in every text they write, translations included. Paste the text below to break the pattern and keep the meaning.
Paste on the left and hit Remove watermark. Takes about ten seconds.Paste above and hit Remove watermark. Takes about ten seconds.
Anthropic's watermark is a version of Google DeepMind's SynthID-Text. As Claude writes, a secret key tilts it toward certain words. Over a few hundred words those choices form a statistical fingerprint that a detector with the key can read. It applies worldwide, and Anthropic is adding it to models launched before August 2026 as well. A translation carries the watermark in full, because every word in it is Claude's.
We swap just enough of the steered words for equivalents to break the pattern, and remove zero-width characters, soft hyphens and markdown leftovers in the same pass. Sentence structure, meaning, length and tone stay the same.
Every Claude model launched since August 2026. Anthropic has said it is adding the watermark to older models too, so treat any recent Claude output as watermarked.
Yes. It lives in the choice of words, not in formatting or hidden characters, so it survives copy-paste and light editing. Only re-choosing the words breaks it.
No. Claude's watermark is a word pattern. A character cleaner leaves it intact. Clean AI Watermark re-chooses the words and strips the hidden characters as well.