Newest papers

A Unified Attack Framework against Statistical Generative Watermarks for Large Language Models

Romina Omidi, Nima Naderloui, Yuan Hong, Binghui Wang · 2026 · arXiv

DOI: 10.48550/arXiv.2610.27653 · open-access PDF

Spanish

El marcado de agua ha surgido como una herramienta clave para identificar el texto generado por los grandes modelos de lenguaje (LLM), con esquemas actuales que incorporan señales estadísticas en el proceso de generación para permitir la detección posterior. En este trabajo, identificamos una debilidad compartida en una amplia clase de métodos de marcado de agua populares de LLM, incluyendo KGW (y sus variantes), GumbelSoft, DipMark, ITS y SynthID. Mostramos que estos esquemas muestran una relación fundamental entre sus puntuaciones de detección y los cambios de probabilidad de token que introducen durante la generación. Este enlace, que llamamos alineamiento de ratio de puntuación, permite un ataque

Machine translation (Meta M2M100); the English original follows.

Abstract

Watermarking has emerged as a key tool for identifying text generated by large language models (LLMs), with current schemes embedding statistical signals into the generation process to enable later detection. In this work, we identify a shared weakness across a broad class of popular LLM watermarking methods, including KGW (and its variants), GumbelSoft, DipMark, ITS, and SynthID. We show that these schemes exhibit a fundamental link between their detection scores and the token probability shifts they introduce during generation. This link, which we call score-ratio alignment, allows an attack

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Is this paper real? This page was built from OpenAlex and arXiv records, not generated. Software can verify any reference in one call: GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.27653, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txt

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange