OOD-MMSafe: Advancing MLLM Safety from Harmful Intent to Hidden Consequences
This work addresses safety issues for robust deployment of autonomous and embodied agents, representing a novel paradigm shift rather than an incremental improvement.
The paper tackles the problem of safety alignment in Multimodal Large Language Models (MLLMs) by shifting focus from harmful intent to consequence-driven safety, introducing the OOD-MMSafe benchmark with 455 query-image pairs that reveals up to 67.5% failure rates in frontier models, and proposes the CASPO framework which reduces failure ratios to 7.3% and 5.7% for specific models.
While safety alignment for Multimodal Large Language Models (MLLMs) has gained significant attention, current paradigms primarily target malicious intent or situational violations. We propose shifting the safety frontier toward consequence-driven safety, a paradigm essential for the robust deployment of autonomous and embodied agents. To formalize this shift, we introduce OOD-MMSafe, a benchmark comprising 455 curated query-image pairs designed to evaluate a model's ability to identify latent hazards within context-dependent causal chains. Our analysis reveals a pervasive causal blindness among frontier models, with the highest 67.5% failure rate in high-capacity closed-source models, and identifies a preference ceiling where static alignment yields format-centric failures rather than improved safety reasoning as model capacity grows. To address these bottlenecks, we develop the Consequence-Aware Safety Policy Optimization (CASPO) framework, which integrates the model's intrinsic reasoning as a dynamic reference for token-level self-distillation rewards. Experimental results demonstrate that CASPO significantly enhances consequence projection, reducing the failure ratio of risk identification to 7.3% for Qwen2.5-VL-7B and 5.7% for Qwen3-VL-4B while maintaining overall effectiveness.