LATIDIA · Ciberseguridad

Suficiencia defensiva en un modelo de seguridad de IA de Stackelberg

arXiv: 2610.09892v1Tipo de anuncio: nuevo Resumen: Los comentarios de las pruebas automatizadas, la formación de equipos humanos y la respuesta a incidentes pueden fortalecer las defensas de un sistema de IA cuando las fallas descubiertas conducen a reparaciones efectivas. Estudiamos

La noticia

arXiv:2610.09892v1 Announce Type: new Abstract: Feedback from automated testing, human red teaming, and incident response can strengthen an AI system's defenses when discovered failures lead to effective repairs. We study when this feedback process provides sufficient protection and when investing in it is economically worthwhile. We begin by showing that an attack surface composed of finite number of inputs is defended with probability 1 if every unresolved attack has a persistent chance of discovery, repairs are effective, and subsequent updates preserve earlier protection. We derive completion-time bounds and extend the analysis to growing attack surfaces, repairs that generalize across related attacks, and multiple discovery mechanisms. These results distinguish eventual protection