LATIDIA · Ciberseguridad

Ediciones mecanicistas certificadas: Garantías de comportamiento para la eliminación y preservación de habilidades

arXiv:2610.03502v1 Tipo de anuncio: cruz Resumen: Las ediciones mecanicistas (ablaciones, ediciones de peso, dirección de activación) son las herramientas estándar para desaprender una capacidad dañina de una red neuronal y al mismo tiempo preservar

La noticia

arXiv:2610.03502v1 Announce Type: cross Abstract: Mechanistic edits (ablations, weight edits, activation steering) are the standard tools for unlearning a harmful capability from a neural network while preserving useful ones. Current approaches validate their effects only by testing, which can never cover an entire continuous region of inputs. Prior work at the interpretability-verification boundary certifies descriptions of a model: what a circuit computes, or whether it faithfully explains the whole. We instead certify the behavioral effect of an edit: that disabling a circuit removes one skill and provably preserves another, for every input in a region; a feature non-interference guarantee in the information-flow-security sense. We demonstrate such certified edits from toy