FULL STORY

Claude's Invisible Watermark Draws Backlash

Anthropic rolled out invisible statistical watermarks in Claude's output for AI content provenance, drawing sharp criticism from John Gruber, who called it a degradation of writing, and Ben Goertzel, who predicted spoofing tools would soon emerge.

2026-08-17 ~ 2026-08-19 · 3 episodes · 9 posts

Episode 1 · Daring Fireball Slams Anthropic's Claude Text Watermarking (2026-08-17, 4 posts)

John Gruber of Daring Fireball harshly criticized Anthropic's text watermarking in Claude, arguing that covertly manipulating word choices during inference distorts the essence of writing despite claims the marks are imperceptible.

Episode 2 · Anthropic Rolls Out Invisible Watermarks for Claude-Generated Text (2026-08-18, 3 posts)

Anthropic introduces invisible watermarks for Claude, embedding machine-readable statistical patterns in token selection that survive copy-paste and light editing, aligning with the EU's upcoming requirement for AI-generated text watermarking from December.

Episode 3 · Goertzel Slams Claude Text Watermarking, Predicts Forgery Tools (2026-08-18, 2 posts)

Ben Goertzel criticized Anthropic's plan to embed invisible statistical watermarks in Claude's output, arguing it mainly serves regulators, schools, and employers. He predicts a market for tools that forge such watermarks if they become a legitimacy standard.