This paper introduces "bears," an ensemble method that enhances the awareness of Neuro-Symbolic (NeSy) models regarding Reasoning Shortcuts (RSs). RSs occur when NeSy models learn concepts with unintended semantics, leading to overconfidence in predictions and compromised reliability. Bears addresses this by making NeSy models uncertain about concepts affected by RSs, enabling users to identify and distrust low-quality predictions. The method does not rely on costly dense supervision, instead using an ensemble of concept extractors trained to encourage uncertainty proportional to RS impact. Experiments show that bears significantly improves RS-awareness across multiple state-of-the-art NeSy architectures, including a high-stakes autonomous driving task, and facilitates the acquisition of informative dense annotations for mitigation. Bears achieves this while maintaining high prediction accuracy and reducing the cost of supervised mitigation through uncertainty-based active learning. The paper also discusses the causes of RSs, existing mitigation strategies, and the benefits of RS-awareness in improving the reliability and interpretability of NeSy models.This paper introduces "bears," an ensemble method that enhances the awareness of Neuro-Symbolic (NeSy) models regarding Reasoning Shortcuts (RSs). RSs occur when NeSy models learn concepts with unintended semantics, leading to overconfidence in predictions and compromised reliability. Bears addresses this by making NeSy models uncertain about concepts affected by RSs, enabling users to identify and distrust low-quality predictions. The method does not rely on costly dense supervision, instead using an ensemble of concept extractors trained to encourage uncertainty proportional to RS impact. Experiments show that bears significantly improves RS-awareness across multiple state-of-the-art NeSy architectures, including a high-stakes autonomous driving task, and facilitates the acquisition of informative dense annotations for mitigation. Bears achieves this while maintaining high prediction accuracy and reducing the cost of supervised mitigation through uncertainty-based active learning. The paper also discusses the causes of RSs, existing mitigation strategies, and the benefits of RS-awareness in improving the reliability and interpretability of NeSy models.