LATIDIA · Ciberseguridad

Tinta invisible, mentiras visibles: cómo las marcas de agua de producción causan alucinaciones en los LLM

arXiv: 2610.04860v1 Tipo de anuncio: nuevo Resumen: La marca de agua de texto ayuda a identificar el contenido generado por IA, pero su efecto sobre la fiabilidad de los hechos sigue siendo poco explorado. En este artículo, estudiamos la alucinación de la marca de agua: hecho

La noticia

arXiv:2610.04860v1 Announce Type: new Abstract: Text watermarking helps identify AI-generated content, but its effect on factual reliability remains underexplored. In this paper, we study watermarking hallucination: factual errors induced or amplified by watermarking even when the required evidence is present in the context and the unwatermarked model can answer correctly. Using a controlled retrieval-augmented generation setting, we compare unwatermarked and watermarked generations under the same context, query, and decoding configuration, and quantify their factual accuracy decrease. Across six representative watermarking methods, including KGW, SWEET, DiPmark, GumbelSoft, Gumbel-Max, and SynthID watermarking, we consistently observe watermark-induced hallucination. Watermarked outputs can remain fluent while introducing factual errors. We attribute this failure mode to