All Claude generated text will have this embedded watermarking. For example, you

Anthropic's text watermark works through patterns in word selection. A secret key influences choices among plausible continuations, allowing a detector to estimate whether Claude participated in producing a passage. The technical explanation says this adds no hidden characters or user-identifying information.

This makes the post's pull-request example conditional. Code offers fewer interchangeable choices, so its watermark is generally sparser; short passages and light proofreading also provide weak evidence. Detection estimates Claude's involvement, without distinguishing original generation from heavy editing. The September 1 update described the detection API as a private preview.