Metraipolis › archive › dc_18f4bb9f712f

Flip, Don't Shuffle: Watermarking LLMs at the Speed of Inference — arXiv:2609.03844

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.03844
License (as archived)
CC0-1.0
Archived
2026-10-17T01:42:10.679Z
SHA-256 of the archived text
45a24cf4fad9e3f463c2b97f59b93a00455094009e2959401f4b19403fcc0435
Stable URL of this page
https://metraipolis.com/documento/dc_18f4bb9f712f

How to cite

Flip, Don't Shuffle: Watermarking LLMs at the Speed of Inference — arXiv:2609.03844 — arXiv (en), https://arxiv.org/abs/2609.03844 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_18f4bb9f712f (Metraipolis archive dc_18f4bb9f712f, sha256 45a24cf4fad9e3f463c2b97f59b93a00455094009e2959401f4b19403fcc0435)
BibTeX
@misc{dc18f4bb9f712f,
  title = {Flip, Don't Shuffle: Watermarking LLMs at the Speed of Inference — arXiv:2609.03844},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.03844},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_18f4bb9f712f; sha256 45a24cf4fad9e3f463c2b97f59b93a00455094009e2959401f4b19403fcc0435},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.03844. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

We introduce Stateless Bernoulli Watermarking (SBW), a new statistical watermark for Large Language Models that determines green list membership through independent per-token Bernoulli trials. Unlike KGW's vocabulary permutation or SynthID's multi-layer tournament, SBW requires only a single comparison per token against a counter-based random number generator, reducing membership complexity to $O(1)$ and enabling single-kernel execution with zero intermediate allocations. We prove that this formulation preserves the same detection guarantees as fixed-size green lists: the z-score test remains $\mathcal{N}(0,1)$ under the null. The stateless architecture enables capabilities unavailable to existing methods: full-vocabulary self-salt watermarking (over 6000$\times$ faster than KGW's self-salt and 2$\times$ faster than SynthID despite biasing the entire vocabulary with candidate-dependent seeding) and architectural compatibility with distributed inference. In end-to-end generation benchmarks, SBW adds less than 1\% overhead at all batch sizes. We additionally identify hash function design as a previously unexplored axis for watermark quality, showing that a GPU-native Jenkins hash improves null calibration by 1.8$\times$ while producing more diverse text. Experiments across two seeding schemes and eight $(\gamma, \delta)$ configurations confirm statistical equivalence with ROC-AUC differences below 0.01.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-04 in cs.CL).
Categorie arXiv: Cryptography and Security (cs.CR) ; Computation and Language (cs.CL); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.03844 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-17T01:11:43.616Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-04?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).