UniSAFE: A Comprehensive Benchmark for Safety Evaluation of Unified Multimodal Models

TL;DR

UniSAFE provides a comprehensive benchmark for evaluating safety risks in unified multimodal models across various tasks and modalities, revealing critical vulnerabilities and emphasizing the need for improved safety alignment.

Contribution

This paper introduces UniSAFE, the first extensive benchmark for system-level safety evaluation of UMMs across multiple modalities and tasks, enabling better safety assessment.

Findings

Current UMMs show significant safety vulnerabilities.

Multi-image and multi-turn tasks are more prone to safety violations.

Image-output tasks are generally more vulnerable than text-output tasks.

Abstract

Unified Multimodal Models (UMMs) offer powerful cross-modality capabilities but introduce new safety risks not observed in single-task models. Despite their emergence, existing safety benchmarks remain fragmented across tasks and modalities, limiting the comprehensive evaluation of complex system-level vulnerabilities. To address this gap, we introduce UniSAFE, the first comprehensive benchmark for system-level safety evaluation of UMMs across 7 I/O modality combinations, spanning conventional tasks and novel multimodal-context image generation settings. UniSAFE is built with a shared-target design that projects common risk scenarios across task-specific I/O configurations, enabling controlled cross-task comparisons of safety failures. Comprising 6,802 curated instances, we use UniSAFE to evaluate 15 state-of-the-art UMMs, both proprietary and open-source. Our results reveal critical vulnerabilities across current UMMs, including elevated safety violations in multi-image composition and multi-turn settings, with image-output tasks consistently more vulnerable than text-output tasks. These findings highlight the need for stronger system-level safety alignment for UMMs. Our code and data are publicly available at https://github.com/segyulee/UniSAFE