Anthropic to watermark all Claude output using invisible SynthID
every · x · 2026-08-23
Anthropic is implementing watermarking for all Claude output using a variant of Google's SynthID technology. This watermark is invisible and undetectable by existing AI detectors.
How it works:
- Uses a secret key to choose between words that were already equally likely, embedding a signal without significantly degrading text quality.
- Preserves the model's probability distribution; the average odds for each word remain the same.
Controversy:
- John Gruber called it “a perversion of writing.”
- Mike from the author team initially feared the tech but changed his mind after understanding it, though he worries that accepting the concept could lead to stronger versions in the future that might actually impact prose.
Interactive Explainer
More from Models
- Flashback: GPT-4 cost $60/M output tokens with 8K context three years ago — gajesh · 2026-08-23
- Open Weights vs Frontier: Just a 3-Point Gap but 1/3 the Cost — MicahBerkley · 2026-08-23
- Ox Alpha Overhyped? Beats GPT-5.6-Luna but Lags Other Frontiers — Al_Grigor · 2026-08-23
- Dev Endorses k3 + NousResearch Harness as Best Combo — markjeffrey · 2026-08-23
- Ornith 1.5 35B Hits 81.8 on GPQA with Thinking Mode, Decodes at 303 tok/s — MikePFrank · 2026-08-23
- Anthropic's Opus 5 and Sonnet 5 flagged as legitimate regressions — bindureddy · 2026-08-23