LATIDIA · Ciberseguridad

Mirage: defensa demostrable contra canales laterales de sincronización de caché LLM KV con aislamiento consciente de la topología

arXiv: 2611.10756v1Tipo de anuncio: nuevo Resumen: los modelos de lenguaje grandes (LLM) se implementan cada vez más como servicios compartidos de múltiples inquilinos, confiando en optimizaciones avanzadas como la caché KV y el uso compartido de prefijos para mejorar

La noticia

arXiv:2611.10756v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly deployed as shared multi-tenant services, relying on advanced optimizations like KV cache and prefix sharing to improve throughput and reduce latency. However, these optimizations introduce a critical vulnerability: attackers can exploit time-to-first-token (TTFT) variations between KV cache hits and misses to infer sensitive prompt tokens via prefix-matching side-channel attacks. Existing defenses indiscriminately restrict cross-tenant cache sharing, which incurs prohibitive performance overhead and negates the fundamental efficiency benefits of the KV cache. We propose Mirage, a topology-aware emission obfuscation framework that provably bounds timing leakage without sacrificing the performance gains of cache sharing. Instead of stalling the GPU or forcing