Anthropic says the new feature won't change how Claude's responses read, its cost to run, or reveal anything about who's using it.

Anthropic will extend support for watermarking AI generations for older models as well.

Claude now watermarks every word it writes. Cold email reply rates already fell to 3.43%, and fully AI-written messages get flagged as spam three times as often.

Charles Hoskinson has released a tool to strip Anthropic’s AI watermarks, while raising broader questions over AI attribution, ownership and the legal implications of Claude’s co-author tags.

On Tuesday, 11 August, Anthropic confirmed a change this column has been expecting for a while: text generated by its newest Claude models now carries an invisible watermark, woven into the words themselves,