Covert Multi-bit LLM Watermarking: An Information Theory and Coding Approach

Sidong Guo 1 , Tyler Kann 1 , Teodora Baluta 2 , Matthieu R. Bloch 1 · 2026

Read Covert Multi-bit LLM Watermarking: An Information Theory and Coding Approach by Sidong Guo 1 , Tyler Kann 1 , Teodora Baluta 2 , Matthieu R. Bloch 1 (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-16709.txt.

Description

Guo and colleagues study multi-bit watermarking for non-autoregressive language models, where the encoder sees token distributions within each block but not causally. They build a theory-based model inspired by diffusion language models and pin down its non-causal capacity exactly, using Gelfand-Pinsker and channel synthesis coding. The hiding is tuned across blocks with a constrained Markov process, and a polar-code algorithm implements it. Tests on LLaDA check error rates and total variation against key randomness.

Catalog record

CREATORSidong Guo 1 , Tyler Kann 1 , Teodora Baluta 2 , Matthieu R. Bloch 1
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41604
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.16709

Browse the full library · About the Tower of Babel