LATIDIA · Ciberseguridad

Aceptar cambios, rechazar ataques: gestión sólida de la memoria para agentes de LLM

arXiv: 2610.41056v1Tipo de anuncio: nuevo Resumen: Dado que los agentes del modelo de lenguaje grande (LLM) dependen de la memoria a largo plazo, lo que almacenan ahora da forma a lo que hacen más adelante. Un administrador de memoria decide cómo cada entrada cambia esta memoria

La noticia

arXiv:2610.41056v1 Announce Type: new Abstract: Since large language model (LLM) agents rely on long-term memory, what they store now shapes what they do later. A memory manager decides how each input changes this memory and which records the agent acts on. It can fail in two ways: keeping a fact that has since changed (outdated memory), or storing a planted passage or injected instruction (memory poisoning). Most prior work studies them separately, yet they call for opposite write decisions, so no fixed write rule handles both. We therefore formalize memory robustness as handling both without doing worse on other inputs. Since accuracy checks only the answer, we measure robustness with