Claude's Hidden AI Label Fades on Short or Edited Text, Complicating New EU Rules

Anthropic confirmed August 14 that Claude tags its AI-written text with a hidden signal that weakens on short messages, edited passages, and code — and disappears after a full rewrite. The EU's new AI law, effective August 2, requires reliable AI-content labeling. The planned detection tool gives yes-or-no answers whose accuracy varies — more like a fading stamp than a permanent one.