LATIDIA · Ciberseguridad

Límites de la Garantía Estadística para la IA Encarnada y Agénica bajo Adversarios Adaptativos

arXiv: 2610.18948v1Tipo de anuncio: nuevo Resumen: Las evaluaciones de seguridad finitas no pueden excluir a un adversario que conserva el comportamiento no atacado hasta la certificación y se activa después. Caracterizamos esta limitación para em

La noticia

arXiv:2610.18948v1 Announce Type: new Abstract: Finite safety assessments cannot exclude an adversary that preserves unattacked behavior until certification and activates afterward. We characterize this limitation for embodied and tool-use AI. An adversary that matches the complete observation law until observable, irrevocable certification and retains intervention access afterward attains $v\mathbb{E}[(n-S_0)_+]$ expected post-certification events over $n$ trials, where $v$ is the active event probability and $S_0$ the unattacked certification time. The result covers adaptive and privately randomized evaluators. Switching at certification is exactly optimal for a fixed likelihood-ratio test under memoryless, event-increasing interventions without a binding intervention budget. A positive clean-certification requirement forces linear worst-case expected excess when intervention raises event rates