Two Birds, One Projection: Harmonizing Safety and Utility in LVLMs via Inference-time Feature Projection

TL;DR

This paper introduces a novel inference-time feature projection method for LVLMs that simultaneously enhances safety and utility by removing bias components, overcoming the traditional safety-utility tradeoff.

Contribution

The authors propose a single-pass feature projection technique that mitigates bias, improving both safety and performance in large vision-language models.

Findings

Effectively breaks the safety-utility tradeoff in LVLMs

Improves safety without degrading reasoning performance

Achieves results with only a single inference pass

Abstract

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.