Cloud file sharing has become a normal part of how people work today. But this convenience brings serious security problems. Traditional antivirus and firewall systems were not built for modern cloud environments where files move quickly between many users and devices. After working with three mid- sized companies for over eight months, we observed that most security teams struggle to detect malicious files shared through Google Drive, SharePoint, and similar platforms. This paper presents a practical AI-based solution that we built and tested in real cloud environments. Our system uses a combination of three machine learning methods — Random Forest, Gradient Boosting, and a small neural network — working together to scan files quickly without slowing down users. We achieved 94.7% accuracy on real-world data, with each file taking less than 100 milliseconds to analyze. The system also explains why it flagged something as suspicious, which helps security people trust the alerts and respond faster. Based on our pilot deployments, organizations saw about 65% fewer successful malware infections through cloud sharing channels