Two Birds, One Projection: Harmonizing Safety and Utility in LVLMs via Inference-time Feature Projection
This addresses the problem of balancing safety against performance for users of LVLMs, offering an efficient solution to a known bottleneck.
The paper tackles the safety-utility tradeoff in Large Vision-Language Models by identifying a modality-induced bias direction that harms both safety and performance, and proposes an inference-time feature projection method to remove this bias, resulting in simultaneous improvements in safety and utility across benchmarks.
Existing jailbreak defence frameworks for Large Vision-Language Models often suffer from a safety utility tradeoff, where strengthening safety inadvertently degrades performance on general visual-grounded reasoning tasks. In this work, we investigate whether safety and utility are inherently antagonistic objectives. We focus on a modality induced bias direction consistently observed across datasets, which arises from suboptimal coupling between the Large Language Model backbone and visual encoders. We further demonstrate that this direction undermines performance on both tasks. Leveraging this insight, we propose Two Birds, One Projection, an efficient inference time jailbreak defence that projects cross-modal features onto the null space of the identified bias direction to remove the corresponding components. Requiring only a single forward pass, our method effectively breaks the conventional tradeoff, simultaneously improving both safety and utility across diverse benchmarks.