LATIDIA · Ciberseguridad

Evaluación de agentes de búsqueda profunda bajo envenenamiento jerárquico de evidencia web

arXiv: 2609.06027v2Tipo de anuncio: reemplazar Resumen: Los agentes LLM aumentados por búsqueda se utilizan cada vez más para las decisiones de los consumidores, lo que los hace vulnerables al envenenamiento por Optimización Generativa de Motores (GEO). Puntos de referencia existentes

La noticia

arXiv:2609.06027v2 Announce Type: replace Abstract: Search-augmented LLM agents are increasingly used for consumer decisions, making them vulnerable to Generative Engine Optimization (GEO) poisoning. Existing benchmarks largely measure whether manipulated content is retrieved or endorsed, but do not track whether an agent verifies suspicious evidence, revises adopted claims, or recovers before producing its final recommendation. We introduce HAE-GEO, a benchmark that tracks the full trajectory from exposure to recovery under progressively more persuasive Web poisoning. Agents interact via a multi-turn Search-Scrape interface across three attack levels (L1 direct assertion, L2 contextual camouflage, and L3 apparent corroboration), supported by a controlled corpus of 72,039 clean pages and 770 poisoned pages per level