Prior work showed that face-recognition systems ingesting RGB images captured via visible-light (VIS) cameras are susceptible to real-world evasion attacks. Face-recognition systems in near-infrared (NIR) are widely deployed for critical tasks (e.g., access control), and are hypothesized to be more secure due to the lower variability and dimensionality of NIR images compared to VIS ones. However, the actual robustness of NIR-based face recognition remains unknown. This work puts the hypothesis to the test by offering attacks well-suited for NIR-based face recognition and adapting them to facilitate physical realizability. The outcome of the attack is an adversarial accessory the adversary can wear to mislead NIR-based face-recognition systems. We tested the attack against six models, both defended and undefended, with varied numbers of subjects in the digital and physical domains. We found that face recognition in NIR is highly susceptible to real-world attacks. For example, ≥96.66% of physically realized attack attempts seeking arbitrary misclassification succeeded, including against defended models. Overall, our work highlights the need to defend NIR-based face recognition, especially when deployed in high-stakes domains.