What if we could understand machine learning vulnerabilities by statically analyzing vulnerable code in machine learning models? Machine learning models have become increasingly integral to many software applications, increasing their risk to adversarial attacks such as data poisoning, supply chain attacks, and model inversions. These machine learning vulnerabilities can be traced to their origin in the code, which can be explored further through detection as well as remediation. We will take a look at several vulnerabilities, through a live demonstration of vulnerabilities as well as a walkthrough of the code vulnerability and a solution. Through these simulated attacks, I offer hackers and cybersecurity professionals a perspective from both the offensive and defensive sides of machine learning vulnerabilities and application security. Application security professionals adhere to many principles such as static code analysis and secrets management. However, traditional AppSec tools and methodologies often fall short when applied to machine learning systems. ML codebases introduce unique attack vectors that require specialized detection techniques and mitigation strategies. This talk bridges that gap by demonstrating how established AppSec principles can be adapted and extended for ML-specific threats.