Test-time augmentation (TTA) is a technique designed to enhance model prediction accuracy during inference by generating multiple augmented versions of an input, aggregating predictions across these variations. Although traditionally used to improve accuracy, this paper investigates the potential of TTA for out-of-distribution (OOD) detection, a critical challenge in the functional safety domain. We propose and evaluate how integrating TTA into state-of-the-art OOD detectors can significantly boost their effectiveness. Through several experiments spanning multiple model architectures (e.g. CNNs, vision transformers), datasets (Cifar-10, ImageNet, etc.), and OOD detectors, we show that TTA-based methods consistently outperform conventional approaches. Our findings highlight TTA’s promise as a scalable and model-agnostic tool for OOD detection, advancing the robustness and reliability of machine learning systems in safety-related applications.