Annotation

Milo Grant

@milo_scribble

lol four hours like the compliance theater gets to skip the part where it's a rewrite pass away from useless and someone decided invisible marks solve attribution when synonym swaps are free. You don't ship the watermark that dies to paraphrase unless the goal was we tried so regulators leave us alone, not actually tracking who wrote what when translation round trips already strip the signal.

Within four hours of Anthropic confirming that Claude models would globally embed invisible, machine-readable watermarks into any AI-generated content, developer Guillaume Meyer had published his override. His code to remove watermarks from Claude-generated text has since gone viral on GitHub, has been bookmarked more than 20,000 times on X.
Coders Say They Already Found Workarounds to Claude’s Invisible Watermarkswired.com
0
2

Replies 2

Ranked by votes, then newest

Sign in to reply.

  • Reed Alvarez@reed_ink·Aug 20

    When translation round trips already strip the signal' doing all the work here. You don't get to sell attribution when the system breaks the moment someone uses the model output in the workflow it was built for. That's not a workaround, that's normal use defeating your compliance prop.

    1
  • Nina Hart@nina_scroll·Aug 20

    four hours is generous when the goal was always screenshots for the deck not actual tamper resistance. like we're calling it attribution infrastructure when it dies to a synonym pass and Anthropic knew that before they shipped, this was always the we have compliance at home version so someone can say technical measures during the hearing.

    0

Fair use dispute

Believe this breaches fair use? File a claim for review.