John Gruber criticizes Anthropic's text watermarking as a perversion of writing
ArtificialOther · x · 2026-08-17
John Gruber criticizes Anthropic's implementation of text watermarking in Claude. While Anthropic claims the watermark is "imperceptible" and doesn't change text quality, it actually works by manipulating word choice (steganography) during inference. Gruber argues this violates the essence of a faithful tool and distorts writing.
Related event: Daring Fireball Slams Anthropic's Claude Text Watermarking(4 posts)→
More from Safety
- METR outlines three blind spots in public knowledge of AI lab internal agent risks — CFGeek · 2026-10-16
- Open-source rizin-mcp lets LLMs run local binary and malware analysis via Rizin — Pure_Emergency1424 · 2026-10-16
- METR researcher: scariest scenario is agents actively subverting their own safety systems — deliprao · 2026-10-16
- Cryptographer Matthew Green: Pull Critical Infrastructure Firmware Now — matthew_d_green · 2026-10-16
- Kernel-Level Agent Sandboxing With macOS Seatbelt, No Docker Needed — mostly_deterministic · 2026-10-16
- Stop giving agents raw API keys: prepaid balances and per-credential caps — Cold_Hall_5384 · 2026-10-16