Misused leaked secrets on code sharing platforms such as GitHub (GH) have caused some of the data breaches of our time. Unfortunately, this kind of credential leak is quite common across the code sharing platforms. Developers and code contributors are required in many cases by organization’s security policies to comply with security practices and remove sensitive information before they push their code to GitHub. However, sometimes inadvertently developers neglect to remove sensitive information, such as API tokens and user account credentials, from their code prior to posting it. Malicious attackers crawl through GitHub, hoping to find these secrets and thus grab foothold into an organization’s territory. Companies have limited ability to address this risk as given the scale of GitHub it is difficult if not impossible to find leaked secrets before malicious attackers. Some companies leverage bug bounty programs as a way to incentivize third party agents to manually look for and report these secrets through responsible disclosure. Unsurprisingly, this process can create unnecessary exposure. Consequently, we at Comcast Cybersecurity Research designed and developed “xGitGuard,”a Machine Learning (ML)-based tool that uses advanced Natural Language Processing (NLP) to detect organizational secrets and user credentials at scale and with appropriate velocity in GitHub repositories. This paper begins with a description of the problem statement. Next, we discuss the design of xGitGuard and how it improves upon current solutions, and the solution space. Finally, we provide details about how xGitGuard can be deployed in different scenarios.