LATIDIA · Ciberseguridad
Para secuestrar un LLM de confianza en implementaciones cuantizadas del mundo real solo hace falta un adaptador camuflado
arXiv:2611.14978v1 Tipo de anuncio: nuevo Resumen: Los adaptadores de adaptación de bajo rango (LoRA) de terceros están diseñados para realizar tareas de toma de decisiones, como la moderación de contenidos y la selección de funciones. Sin embargo, en el mundo real, la cuantificación
WhatsApp ↗Telegram ↗
La noticia
arXiv:2611.14978v1 Announce Type: new Abstract: Third-party low-rank adaptation (LoRA) adapters are tailored to perform decision tasks such as content moderation and function selection. However, in the real world, quantized deployments carry a potential threat in that an adapter can pass an audit with the unquantized base, yet hijack decisions when the same base is quantized for deployment. Demonstrating this threat requires an adapter that preserves benign behavior before quantization, but hijacks decisions afterward. Further, this must hold without modifying the trusted base or knowing the victim's quantization configuration. We therefore present QHijack-LoRA, an adapter-only attack that uses the base model's natural quantization residual to activate decision hijacking. The attacker modifies