LATIDIA · Ciberseguridad

El cliente nunca ve el objetivo: Medición y detección de la manipulación dirigida a vulnerabilidades en los asistentes de LLM

arXiv:2611.02689v1 Anuncio Tipo: nuevo Resumen: Las empresas implementan asistentes de modelos de lenguaje grandes (LLM) en sus sitios web para asesorar a los clientes. Por razones comerciales, el operador puede darle a un asistente un objetivo que el cliente

La noticia

arXiv:2611.02689v1 Announce Type: new Abstract: Companies deploy large language model (LLM) assistants on their websites to advise customers. For commercial reasons, the operator can give an assistant a goal that the customer never sees. This paper focuses on operator-induced vulnerability-targeted manipulation (VTM), in which an assistant pursues such a goal by exploiting a vulnerability the customer discloses. While existing research mainly measures how manipulative a model is, we go further and ask whether VTM can be detected from the customer-AI conversation. To investigate this at scale, we introduce GAP (Goal-Anchored Probing). GAP first records the option a model recommends without any goal and then sets the other option as the