Daring Fireball · 2026-08-16 · notable
John Gruber — Claude's text watermark is 'a perversion of writing'
John Gruber argues that Anthropic's new text watermark damages what Claude writes. A secret key nudges Claude's word choices, so Gruber says a reader can no longer tell whether a word was picked for the sentence or for the mark.
Gruber's case against Claude's text watermark: it changes your words for someone else's benefit.
What is it?
Claude's new text watermark is the target of a sharp critique from John Gruber on Daring Fireball. Gruber accepts that the mark is invisible, but rejects Anthropic's claim that it leaves quality untouched. His objection is that the system steers Claude toward some words over others for a reason that has nothing to do with the person reading them.
How does it work?
The watermark works by biasing the pick between words the model rates as roughly equal, using a secret key. Gruber's answer is that 'roughly equal' is the writer's judgment to make — choosing 'grey' over 'overcast' is a real decision, not a coin flip. He adds two more complaints: only Anthropic can read the mark, and by his reading it covers any output longer than 200 tokens, including private chats and proofreading where nobody is being misled.
Why does it matter?
Daring Fireball is widely read by developers and writers, so this piece gives shape to a complaint many Claude users have raised since the watermark was announced. Every large lab now faces the same EU AI Act rule that pushed Anthropic here, and Gruber's argument sets out the cost side of it: transparency for the public, paid for with word choices the user did not make.
Who is it for?
writers and developers who publish Claude output