LATIDIA · Ciberseguridad

Protección del conocimiento sensible en el desaprendizaje de los LLM: la amenaza de los «proxies» y una defensa consciente de la superposición

arXiv:2611.15013v1 Tipo de anuncio: nuevo Resumen: El desaprendizaje automático permite al propietario de grandes modelos de lenguaje (LLM) eliminar entidades específicas de un modelo entrenado sin tener que volver a entrenarlo desde cero. El desaprendizaje automático existente

La noticia

arXiv:2611.15013v1 Announce Type: new Abstract: Machine unlearning lets the owner of large language models (LLMs) remove specific entities from a trained model without retraining from scratch. Existing machine unlearning methods focus on removing the requested entities while maintaining utility. However, we observe that the representations of distinct entities in an LLM may overlap in shared parameter subspaces, a phenomenon known as knowledge superposition. Removing one entity may then affect other entities. This exposes a new security threat: legitimate unlearning requests can be exploited to remove sensitive entities that were never requested. This threat is unlike data poisoning or backdoor attacks. We define the attack as the Proxy Unlearning Attack (PUA),