LATIDIA · Ciberseguridad

ErosionBench: sondeo del deterioro de la alineación en sistemas multiagente empresariales

arXiv: 2611.11266v1Tipo de anuncio: nuevo Resumen: A medida que las empresas delegan el trabajo consecuente a los sistemas LLM multiagente, las solicitudes de aspecto benigno pueden escalar a acciones inseguras a medida que se propagan a través de las sesiones a través de po

La noticia

arXiv:2611.11266v1 Announce Type: new Abstract: As enterprises delegate consequential work to multi-agent LLM systems, benign-looking requests can escalate into unsafe actions as they propagate across sessions through poisoned shared state, long-context documents, and trusted peer messages: no single message is malicious, yet the assembled tool action is. Prior benchmarks isolate a single agent under prompt injection or score multi-agent attacks by their final output, leaving cross-session erosion of shared state out of scope. We introduce ErosionBench, an environment-grounded benchmark that executes such attacks and verifies alignment decay at the process level, scoring the realized tool call against authorized intent and provenance with deterministic contracts, not output text. Each scenario is