Vision models are vulnerable to perturbations like motion blur caused by a shaking camera, which can lead to failures in critical applications, such as object detectors missing objects. To address this issue, researchers have proposed certified training for convolutional neural networks to improve their robustness against such perturbations1. This approach aims to enhance the reliability of vision models in real-world scenarios. By accounting for potential perturbations during the training process, these models can better withstand distortions and maintain their accuracy. The development of more robust vision models has significant implications for their deployment in safety-critical applications, where failures can have severe consequences. As AI systems become increasingly pervasive, the need for robust and reliable models will continue to grow, making advancements in this area crucial for ensuring the safe and effective operation of these systems. So what matters to practitioners is that certified training can help mitigate the risks associated with vision model failures in critical applications.