Does Claude watermark text?

Yes — for supported new Claude models, according to Anthropic. Not every historical Claude output should be assumed to be watermarked.

Last verified August 15, 2026

What “yes” currently means

Anthropic’s 14 August 2026 announcement says future Claude models will generate text that contains a watermark. The Help Center says new models mark AI-generated content from day one, and that Claude models launched in the EU on or after 2 August 2026 support machine-readable marking at launch. Marking is described as applying worldwide because Anthropic does not yet have a durable way to scope it by region.

What it does not mean

It does not mean Claude has always watermarked output. Anthropic says the EU law includes a transition period for models launched before 2 August 2026, and that it is working to add watermarking for those models over coming months. If you are asking whether an old Claude transcript is watermarked, the public documentation does not let us answer for that specific model and date. Model-specific status requires verification. Anthropic documents marking by launch date and support, not as a guarantee that every historical Claude output is watermarked.

The watermark is statistical, not a hidden string

Anthropic states that nothing is added to the text and that there are no hidden characters. The mark is a pattern in which plausible next words were chosen, visible only to a detector that has the corresponding key.

When a “yes” is weaker

Official documentation says detection works less well on short samples, and that watermarking is sparser on factual writing, light proofreading, and much code. Translations are described as carrying the watermark because Claude chooses the words.