Despite their rapid advancement, Multimodal Large Language Models (MLLMs) remain vulnerable to diverse safety risks. Current benchmarks fail to provide reliable assessments due to limited risk coverage, insufficient scale, and the oversight of complex modality combinations (e.g., cross-modal risks). To address this, we introduce the Unified Safety Benchmark (USB), a comprehensive framework covering 61 risk categories across four distinct modality interactions. We first demonstrate that existing benchmarks—even when aggregated—leave significant coverage gaps. To bridge this, we design a sophisticated data synthesis pipeline that generates complementary data, ensuring balanced coverage across all risk dimensions. Furthermore, beyond evaluating vulnerability to harmful queries, USB incorporates the simultaneous assessment of model over-refusal on benign inputs as an integrated diagnostic suite. Experimental results, evaluating 22 MLLMs across 244 risk-modality intersections, demonstrate that existing MLLMs still struggle with the trade-off between avoiding vulnerabilities and over-refusal. Models are particularly vulnerable to image-only or cross-modal risky inputs, highlighting the persistent need for refined safety mechanisms. Warning: This paper contains unfiltered and potentially harmful content that may be offensive.

As Large Language Models (LLMs) become increasingly influential in reasoning tasks, ensuring their trustworthiness and introspective self-awareness is critical. This research introduces the Think-Solve-Verify (TSV) framework, an innovative strategy tailored to explore LLMs’ trustworthiness, introspective self-awareness, and collaborative reasoning. This method accentuates a model’s capability to construct introspective reasoning processes from answers and ensure their trustworthiness. The reasoning with TSV consistently performs at or near the top across the majority of datasets with a single interaction with LLM. Moreover, we refine the voting process of self-consistency within the Chain-of-Thought (CoT) approach, leading to notable accuracy enhancements. In our evaluations, this approach improved performance from 67.3% to 72.8% on the AQuA dataset. Furthermore, we delve into the model’s ability to explain the given answers, highlighting the significance of discerning genuine comprehension from mere guesswork.