LATIDIA · Ciberseguridad
Olvídese del linaje: borrado criptográfico con seguridad prospectiva y linaje completo para la memoria de los agentes de LLM
arXiv:2611.18133v1 Tipo de anuncio: nuevo Resumen: Los agentes de modelos de lenguaje grande (LLM) mantienen una memoria a largo plazo que el propio modelo escribe: resúmenes, datos extraídos e incrustaciones derivadas de lo que dijeron los usuarios. Eliminar una m
WhatsApp ↗Telegram ↗
La noticia
arXiv:2611.18133v1 Announce Type: new Abstract: Large language model (LLM) agents keep a long-term memory that the model itself writes: summaries, extracted facts and embeddings derived from what users said. Deleting a message is therefore not enough: its derivatives, storage residue and backups survive. We define lineage-complete forward-secure erasure: after a forget request, an adversary that later obtains all storage and the agent's current keys learns nothing about the forgotten input or any memory derived from it. We present LinEra, which encrypts each memory under a key derived from its parents' keys along the derivation graph; source keys come from a puncturable pseudorandom function, so forgetting is a single puncture and