LATIDIA · Ciberseguridad

De los ataques A2a a la defensa de capa envolvente: evaluación de equipos rojos de agentes LLM y un modelo de defensa de ataque isomórfico de tres capas

arXiv: 2610.00392v1Tipo de anuncio: nuevo Resumen: Los protocolos de interacción de agentes como ACP y A2a han movido a los agentes basados en LLM hacia la colaboración multiagente, introduciendo nuevas amenazas de seguridad. Una tarea enviada por un par remoto

La noticia

arXiv:2610.00392v1 Announce Type: new Abstract: Agent interaction protocols such as ACP and A2A have moved LLM-based agents toward multi-agent collaboration, introducing new security threats. A task sent by a remote peer over A2A is treated as a legitimate request, providing a natural channel for indirect prompt injection. Existing agent security evaluations mostly rely on a single metric, the attack success rate (ASR), and cannot distinguish whether an attack failed because the LLM recognized the malicious content or because a mechanism at the agent layer blocked execution. To address this, we propose A2A-TIBA, an attack principle combining indirect prompt injection with bypass circumvention. Through implant-command-exfiltration steps, it induces the target agent