RuleSafe-VL: Evaluating Rule-Conditioned Decision Reasoning in Vision-Language Content Moderation
For researchers and practitioners in content moderation, the benchmark provides a diagnostic tool to assess whether models apply policy rules correctly rather than relying on superficial cues.
The paper introduces RuleSafe-VL, a benchmark for evaluating rule-conditioned decision reasoning in vision-language content moderation, revealing that current VLMs struggle with rule-relation recovery (best model 64.8 Macro-F1) and decision-state prediction (64.5 Macro-F1), with some safety-oriented models scoring below 7 Macro-F1.
Platform content moderation applies explicit policy rules and context-dependent conditions to decide whether user content is allowed, restricted, or removed. A correct moderation outcome must therefore depend on which rules a case activates, how those rules interact, and whether the available evidence is sufficient. Current multimodal safety benchmarks largely reduce moderation to matching predefined final labels, leaving this underlying rule structure untested. As a result, a high benchmark score reveals little about whether a model applies the policy correctly or arrives at the correct label through superficial cues. To evaluate this rule-governed process, we introduce RuleSafe-VL, a benchmark for rule-conditioned decision reasoning in vision-language content moderation. Derived from publicly available platform moderation policies, RuleSafe-VL formalizes 93 atomic rules and 92 typed rule relations, yielding 2,166 context-sensitive image-text cases across three high-risk policy families. Its four diagnostic tasks decompose moderation into a rule-conditioned decision chain. They identify activated rules, recover rule interactions, judge decision sufficiency, and resolve outcomes once missing context is supplied. Experiments on 10 frontier, open-source, and safety-oriented VLMs reveal rule-relation recovery as the dominant bottleneck, where the best model reaches only 64.8 Macro-F1 and some safety-oriented models fall below 7 Macro-F1. Decision-state prediction also remains unreliable, peaking at 64.5 Macro-F1. RuleSafe-VL shifts moderation evaluation from final-label scoring toward diagnostic assessment of rule-conditioned decision reasoning.