Interactive Demo Explains Anthropic's Watermarking via Probability Bias
VeryWellVersed · x · 2026-08-21
Ian Lurie published an interactive demo explaining the likely mechanism behind Anthropic's watermarking. LLMs select words based on probability lists; watermarking subtly tweaks this list to bias the likelihood of certain words. Detectors analyze the text for this specific pattern shift (a 'green list' of words) to determine if it was generated by Claude. The demo lets users adjust bias intensity and observe changes in word probabilities and detection metrics like z-score.
More from Safety
- New paper uses revealed-preference methods to unpack why AI agents cooperate — brwilder · 2026-10-16
- Anthropic gives METR employee-level access, waives right to block its risk assessments — TechCrunch AI · 2026-10-16
- Berkeley scholars: concentrated chip supply chain makes a hardwired frontier AI pause feasible — wfithian · 2026-10-16
- Anthropic's GRAFT routes flagged gradients into a deletable adapter, cutting reward hacking to near zero — rohanpaul_ai · 2026-10-16
- Polymarket opens 50/50 market on more states banning Flock police cameras by end of 2026 — Polymarket · 2026-10-16
- Reproducing the Hugging Face security incident: a technical deep-dive, Part I — jessehu · 2026-10-16