<p>Video anomaly detection remains a challenging problem, particularly in weakly supervised settings where only video-level annotations are available. In this paper, we present CLIP with Adapters for Robust weakly supervised video Anomaly detection (CLARA), a lightweight and effective framework that adapts the Contrastive Language–Image Pretraining (CLIP) model for practical video anomaly detection. CLARA incorporates three complementary, parameter-efficient modules: a Spatial Adapter, an MLP Adapter, and an Augmented Temporal Adapter. The Spatial Adapter employs prompt-guided spatial attention to emphasize anomaly-relevant regions while suppressing background redundancy. The MLP Adapter mitigates training–testing distribution shifts under weak supervision through cosine-similarity-based feature retrieval, improving generalization. The Augmented Temporal Adapter enhances temporal modeling by introducing a Local Dynamics Enhancer, which explicitly captures short-term temporal irregularities and high-frequency dynamics that are often overlooked by conventional temporal encoders. By jointly improving spatial discrimination, temporal sensitivity, and robustness to weak supervision, CLARA effectively adapts CLIP to video anomaly detection with minor additional parameters and computational overhead. Extensive experiments on three widely used benchmarks—UCF-Crime, XD-Violence, and ShanghaiTech—demonstrate that CLARA consistently achieves competitive or superior performance compared with state-of-the-art methods.</p>