LATIDIA · Ciberseguridad

BiShield-TEE: Sobre la (in)seguridad de la ofuscación de peso unilateral en la partición LLM blindada con TEE en el dispositivo

arXiv: 2610.21565v1Tipo de anuncio: nuevo Resumen: La implementación en el dispositivo de modelos de lenguaje grandes (LLM) se adopta cada vez más debido a los requisitos de latencia y privacidad, pero expone los modelos patentados a la ingeniería inversa

La noticia

arXiv:2610.21565v1 Announce Type: new Abstract: On-device deployment of large language models (LLMs) is increasingly adopted due to latency and privacy requirements, but exposes proprietary models to reverse engineering and intellectual property theft. Trusted Execution Environments (TEEs) provide hardware-enforced isolation for on-device inference, yet most TEE-based parameter obfuscation relies on unilateral matrix transformations that obfuscate only output side. We identify a vulnerability in this design, termed the \emph{Solution Determinacy Property (SDP)}: the obfuscation secret is uniquely determined, exposing the structure of the true key and enabling attackers to break the protection. Under our attacks, representative methods \textsc{GroupCover} and \textsc{ArrowCloak} are reduced to negligible protection, with attack accuracy reaching over $2.4\times$ the