LATIDIA · Ciberseguridad
¿Hasta qué punto son diversas las salvaguardas por capas? Parámetros de estimación y medición para conjuntos de defensas contra el «jailbreaking» de los LLM
arXiv:2608.28327v2 Tipo de anuncio: sustitución Resumen: Los profesionales protegen los grandes modelos de lenguaje (LLM) superponiendo medidas de seguridad: un filtro de probabilidad, un clasificador de protección, sondas de estado oculto y una intervención en el momento de la decodificación.
WhatsApp ↗Telegram ↗
La noticia
arXiv:2608.28327v2 Announce Type: replace Abstract: Practitioners protect large language models (LLMs) by stacking safeguards: a likelihood filter, a guard classifier, hidden-state probes, a decoding-time intervention. Such a stack is an ensemble of learned detectors, and ensemble theory states when combination pays: when members fail on different inputs. Stacking is recommended throughout the LLM security literature, but this condition has not been measured for any pair of safeguards. The condition is three quantities, not one, and conflating them is why it has resisted measurement. Separately optimized attacks license only a claim about which \emph{behaviors} are jointly vulnerable; the identity that governs a deployed stack is written over shared \emph{inputs}; and neither