Anthropic Watermarks All Claude Output Globally, Including User-Edited Text
Anthropic has begun embedding invisible watermarks into all text produced by Claude, including content that users wrote themselves but asked Claude to edit or proofread. The measure, announced in August 2026, is tied to Anthropic's commitment to comply with Article 50 of the EU AI Act, which requires AI-generated text to be detectable as machine-made. The watermark works by subtly biasing Claude's token selection at each step, encoding a hidden signal that is statistically detectable by those holding the cryptographic key. Crucially, because Claude regenerates text token by token even during light edits, any human-authored content it touches also receives the watermark. Anthropic has applied this policy globally, not just in EU jurisdictions, meaning users worldwide cannot opt out, view, or remove the embedded signal.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in