CrowdStrike employs an innovative approach to enhance malware detection in PowerShell scripts by utilizing a character-level convolutional neural network (charCNN) in its Kestrel model. This methodology leverages machine learning to analyze the raw input of scripts, bypassing traditional feature-engineering steps and enabling better detection of malicious content by examining character patterns. The charCNN is lightweight and offers improved interpretability, as its convolutional layers maintain spatial patterns that aid in identifying malware. Initial experiments with different classifiers did not meet expectations, but the charCNN, trained with an augmented alphabet of nearly 5,000 characters, demonstrated promising results, significantly reducing false positives while maintaining a high detection rate of malware. This advancement highlights CrowdStrike's commitment to applying sophisticated ML techniques to cybersecurity challenges, aiming to improve malware detection in scripting environments.