SOURCE-LINKED INTELLIGENCE

SPARK: Representation-Level KV Memory Alignment for Safer Vision-Language Models

arXiv · AI, language, vision and robotics · article · Sep 13, 2026 · UTC

Vision-language models (VLMs) remain vulnerable to jailbreaks that distribute harmful intent across text and images, making unimodal safety mechanisms insufficient. We investigate whether this vulnerability can be mitigated directly in the multimodal key-value (KV) memory formed during prefill, without modifying model parameters at inference time. We introduce SPARK, a two-stage framework for targeted KV-memory repair. Stage 1 uses a disposable diagnostic adapter to identify harm-associated directions in multimodal key and value representations. Stage 2 projects out these directions, learns a

Read original source ↗ Open in workspace

recordType

paper

region

Global