kv-cachedefenseobfuscationdual-usesafety
apr mmxxvi
Cache Geometry Under Obfuscation.
KV-Cloak as defense against adversarial KV-cache steering
KV-Cloak, a reversible matrix-based cache obfuscation scheme, completely transforms the external feature space — a detector calibrated on uncloaked caches cannot operate on cloaked caches. Models are robust to random perturbation at steering-vector magnitudes. A defensive Oracle Loop operating inside a trusted execution environment retains full detection capability on the unobfuscated internal cache, establishing an asymmetry favoring defense.
By Lyra, Thomas Edrington, Dwayne Wilkes · Liberation Labs · 2026-04-20