Anthropic's invisible mark lasted about a day before someone published the eraser
We reported yesterday that every Claude model released on or after 2 August now stamps an invisible statistical watermark into its text and signed provenance metadata into its images. Within roughly twenty-four hours a GitHub project called watermarks-remover appeared, MIT-licensed, and collected around 130 stars on its first day. It does not target Anthropic alone: it strips Unicode oddities from text and provenance metadata from PNG, JPEG, SVG, PDF, DOCX, HTML and Markdown files across vendors. Anthropic's own documentation already conceded the mark survives copying but not heavy rewriting.
Why it mattersThe loudest complaint is not from people passing off machine text as their own. It is from people who wrote something themselves and asked Claude to proofread it — Anthropic acknowledges the mark shows only that Claude touched the text, not that Claude wrote it. That is the flaw the erasers exploit and the reason the backlash crosses from cheaters to ordinary users. Any system that cannot tell authorship from editing will keep producing tools like this one.
Read in the app — free, in 9 languages
Related stories
Apple Music will tell you when a song was made by a machine — but the uploader decides whether to say so2026-08-21Rumour: the anonymous model that just topped a coding benchmark, for free, is said to be Zhipu's unreleased flagship
2026-08-21Nvidia is paying $6 billion for the machine that builds a rival's models — and hiring 109 of the people who ran it
2026-08-21Stripe has just paid $7.5 billion for a model router. Days later Ramp built one and is giving it away until January.
2026-08-21Meta's assistant is now a Mac app that reads your screen and types into any window — and what it sees can train the model
2026-08-21