Anthropic explains Claude's text watermarking: using AR decoding for tie-breaking
prdeepakbabu · x · 2026-08-15
Anthropic published an article detailing the text watermarking technology for Claude models. The method embeds watermarks by using a random number generator (based on a secret key) to perform "tie-breaking" among candidate words with similar probabilities during autoregressive decoding. This approach adds no hidden characters, does not affect output quality or cost, and cannot be traced back to specific users. Anthropic stated this is in response to requirements from the EU AI Act.
Related event: Anthropic Deploys Text Watermarking for Claude to Comply with EU AI Act(31 posts)→
More from Models
- Using System One models in Swift: fast, deterministic decisions via Apple Foundation Models — rxwei · 2026-10-03
- Linux Kernel CVEs Surge From ~500 to 1500+ Per Release, LLMs Blamed for Bulk of the Rise — burny_tech · 2026-10-03
- Developer Complains OpenAI's Coding Model Endlessly Scopes Creeps Instead of Finishing Tasks — DavidWells · 2026-10-03
- Sonnet 5 Spotted in Google Antigravity Backend, Which Still Runs Sonnet 4.6 — brandon_galang · 2026-10-03
- Post-training Yandex AliceAI-80B-A3B from scratch: a NaN bug in custom V100 kernels killed one run — jjusko20 · 2026-10-03
- Every's Dev Day chat with Matthew Berman: budget gone the moment he tried Ultrafast — every · 2026-10-03